《墾?。?024生成式人工智能海外合規白皮書-東南亞篇（英文版）（172頁）.pdf》由會員分享，可在線閱讀，更多相關《墾?。?024生成式人工智能海外合規白皮書-東南亞篇（英文版）（172頁）.pdf（172頁珍藏版）》請在三個皮匠報告上搜索。

1、Generative AI-Overseas Compliance White PaperKinding Law FirmKinding(Singapore),WEEE Consulting,Booster hubAugust 2024Southeast AsiaCONTENTSABOUT THE AUTHORDISCLAIMERPREFACEI.Singapore(I)Generative AI Industry and Regulation in Singapore1.Current status and position of generative AI in Singapore2.Re

2、gulatory framework for generative AI in Singapore(II)Core Issues of Generative AI Compliance in Singapore1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and data cross-border4.Developer securit

3、y responsibility5.Content security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionII.VIETNAM(I)Generative AI Industry and Regulation in Vietnam1.Current status and position of generative AI in Vietnam2.Regulatory framework for

4、generative AI in Vietnam(II)Core Issues of Generative AI Compliance in Vietnam1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content s

5、ecurity6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionIII.THAILAND(I)Generative AI Industry and Regulation in Thailand1.Current status and position of generative AI in Thailand2.Regulatory framework for generative AI in Thaila

6、nd(II)Core Issues of Generative AI Compliance in Thailand1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrigh

7、tability of intellectual property of generated content7.Transparency of large model8.User rights protectionIV.MALAYSIA(I)Generative AI Industry and Regulation in Malaysia1.Current status and position of generative AI in Malaysia2.Regulatory framework for generative AI in Malaysia(II)Core Issues of G

8、enerative AI Compliance in Malaysia1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellectu

9、al property of generated content7.Transparency of large model8.User rights protectionV.INDONESIA(I)Generative AI Industry and Regulation in Indonesia1.Current status and position of generative AI in Indonesia2.Regulatory framework for generative AI in Indonesia(II)Core Issues of Generative AI Compli

10、ance in Indonesia、1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellectual property of ge

11、nerated content7.Transparency of large model8.User rights protectionVI.PHILIPPINES(I)Generative AI Industry and Regulation in the Philippines1.Current status and position of generative AI in the Philippines2.Regulatory framework for generative AI in the Philippines(II)Core issues of Generative AI Co

12、mpliance in the Philippines1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellectual prope

13、rty of generated content7.Transparency of large model8.User rights protectionVII.KINDINGS EXPERIENCE IN GOING GLOBAL WITH AI1 CONTENTS1229131721263439434613II.VIETNAM(I)Generative AI Industry and Regulation in Vietnam1.Current status and position of generative AI in Vietnam2.Regulatory framework for

14、 generative AI in Vietnam(II)Core Issues of Generative AI Compliance in Vietnam1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content

15、security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionIII.THAILAND(I)Generative AI Industry and Regulation in Thailand1.Current status and position of generative AI in Thailand2.Regulatory framework for generative AI in Thail

16、and(II)Core Issues of Generative AI Compliance in Thailand1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrig

17、htability of intellectual property of generated content7.Transparency of large model8.User rights protectionIV.MALAYSIA(I)Generative AI Industry and Regulation in Malaysia1.Current status and position of generative AI in Malaysia2.Regulatory framework for generative AI in Malaysia(II)Core Issues of

18、Generative AI Compliance in Malaysia1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellect

19、ual property of generated content7.Transparency of large model8.User rights protectionV.INDONESIA(I)Generative AI Industry and Regulation in Indonesia1.Current status and position of generative AI in Indonesia2.Regulatory framework for generative AI in Indonesia(II)Core Issues of Generative AI Compl

20、iance in Indonesia、1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellectual property of g

21、enerated content7.Transparency of large model8.User rights protectionVI.PHILIPPINES(I)Generative AI Industry and Regulation in the Philippines1.Current status and position of generative AI in the Philippines2.Regulatory framework for generative AI in the Philippines(II)Core issues of Generative AI C

22、ompliance in the Philippines1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellectual prop

23、erty of generated content7.Transparency of large model8.User rights protectionVII.KINDINGS EXPERIENCE IN GOING GLOBAL WITH AI2 CONTENTS5760636770727373735076777782858557515154and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer

24、 security responsibility5.Content security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionIV.MALAYSIA(I)Generative AI Industry and Regulation in Malaysia1.Current status and position of generative AI in Malaysia2.Regulatory fra

25、mework for generative AI in Malaysia(II)Core Issues of Generative AI Compliance in Malaysia1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibili

26、ty5.Content security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionV.INDONESIA(I)Generative AI Industry and Regulation in Indonesia1.Current status and position of generative AI in Indonesia2.Regulatory framework for generativ

27、e AI in Indonesia(II)Core Issues of Generative AI Compliance in Indonesia、1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content secur

28、ity6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionVI.PHILIPPINES(I)Generative AI Industry and Regulation in the Philippines1.Current status and position of generative AI in the Philippines2.Regulatory framework for generative

29、AI in the Philippines(II)Core issues of Generative AI Compliance in the Philippines1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Cont

30、ent security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionVII.KINDINGS EXPERIENCE IN GOING GLOBAL WITH AI3 CONTENTS869094959697991001011011041081081081101151151181188.User rights protectionV.INDONESIA(I)Generative AI Industry

31、 and Regulation in Indonesia1.Current status and position of generative AI in Indonesia2.Regulatory framework for generative AI in Indonesia(II)Core Issues of Generative AI Compliance in Indonesia1.Relationship and characterization between foundation models and AI products2.Compliance of large model

32、 pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionVI.PHILIPPINES(I)Generative AI Industry and Regulation

33、 in the Philippines1.Current status and position of generative AI in the Philippines2.Regulatory framework for generative AI in the Philippines(II)Core issues of Generative AI Compliance in the Philippines1.Relationship and characterization between foundation models and AI products2.Compliance of la

34、rge model pre-training using public data3.Data localization and cross-border data4.Developer security responsibility5.Content security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionVII.KINDINGS EXPERIENCE IN GOING GLOBAL WITH

35、AI4 CONTENTS119121142143143143122122130132132134134137137139140140(II)Core issues of Generative AI Compliance in the Philippines1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and cross-border

36、data4.Developer security responsibility5.Content security6.The copyrightability of intellectual property of generated content7.Transparency of large model8.User rights protectionCONTACT US5 CONTENTS148148148150153153155156156159ABOUT THE AUTHORKinding Law FirmKinding Law Firm was established in July

37、 2017 by a group of lawyers with rich experience in cyberlaw practice.It is a professional law firm specializing in delivering cyberlaw legal services to its clients.Clients that Kinding lawyers are serving or have served include,but are not limited to:Tencent,Xiaomi,Ant International,AliGenie,Baidu

38、,360,Meizu,YY,Vipshop,Focus Technology,and other integrated and niche market Internet companies.It also includes Johnson&Johnson,Nike,Panasonic,Tesla(China),Smart Automobile,SAIC Volkswagen,OPPO,Hitachi(China),Transsion Holdings,ENEROC New Energy and other intelligent manufacturing and emerging prod

39、uct markets.Kindings practice covers platform governance,product compliance,investment and M&A,data compliance,intellectual property rights,cross-border dispute resolution,etc.,while the industry sectors cover integrated web platforms,AIGC,NFT/blockchain,digital entertainment,and smart manufacturing

40、.WEEE Consulting WEEE Consulting is committed to empowering companies to navigate cross-cultural challenges in global expansion.Founded in 2022,WEEE Consulting offers a comprehensive range of professional services,including market entry assessment,brand building,global marketing&PR,and cross-cultura

41、l leadership training.The founding team is comprised of professionals with extensive experience in global markets,having previously worked at large multinational technology companies.With a deep understanding of diverse cultures,market dynamics,regulatory landscapes,and industry trends,WEEE Consulti

42、ng has successfully served a wide range of clients,from global giants to emerging startups.The firm tailors effective brand marketing strategies and delivers cross-cultural training,enabling globally-minded,ambitious companies to anchor themselves in a complex,ever-changing environment and achieve g

43、reater success.Booster Hub Booster Hub is an innovative commercialisation base with globalised services and trade as its industrial core,and is the Global Cross-border E-commerce DTC Brand Innovation Centre of China Cross-Border E-Commerce Comprehensive Pilot Zone.Booster Hub focuses on introducing,

44、incubating and serving enterprises with innovation and industrial advantages in the direction of the global market,and building an industrial ecosystem of large,medium,small and medium-sized chains around the four core service directions of global business flow,logistics,capital flow and information

45、 flow.In addition,support will be provided in the form of vertical market-based industrial fund support,industry media matrix services,and cross-border service provider system docking.Drafters:Kinding Singapore Office(In alphabetical order)Melody Chen,Tate He,Aria Hu,Shuai Huang,Joy Huang,Zhenghao J

46、iang,Matt Ma,Mou Ruotao,Serena Shu,Nathan Xia,Junchao Zhu,Sarah Zhu，Charlotte MUAdvisor:Shirley Yao,Wayne Luo 1 ABOUT THE AUTHERWEEE Consulting WEEE Consulting is committed to empowering companies to navigate cross-cultural challenges in global expansion.Founded in 2022,WEEE Consulting offers a comp

47、rehensive range of professional services,including market entry assessment,brand building,global marketing&PR,and cross-cultural leadership training.The founding team is comprised of professionals with extensive experience in global markets,having previously worked at large multinational technology

48、companies.With a deep understanding of diverse cultures,market dynamics,regulatory landscapes,and industry trends,WEEE Consulting has successfully served a wide range of clients,from global giants to emerging startups.The firm tailors effective brand marketing strategies and delivers cross-cultural

49、training,enabling globally-minded,ambitious companies to anchor themselves in a complex,ever-changing environment and achieve greater success.Booster Hub Booster Hub is an innovative commercialisation base with globalised services and trade as its industrial core,and is the Global Cross-border E-com

50、merce DTC Brand Innovation Centre of China Cross-Border E-Commerce Comprehensive Pilot Zone.Booster Hub focuses on introducing,incubating and serving enterprises with innovation and industrial advantages in the direction of the global market,and building an industrial ecosystem of large,medium,small

51、 and medium-sized chains around the four core service directions of global business flow,logistics,capital flow and information flow.In addition,support will be provided in the form of vertical market-based industrial fund support,industry media matrix services,and cross-border service provider syst

52、em docking.Drafters:Kinding Singapore Office(In alphabetical order)Melody Chen,Tate He,Aria Hu,Shuai Huang,Joy Huang,Zhenghao Jiang,Matt Ma,Mou Ruotao,Serena Shu,Nathan Xia,Junchao Zhu,Sarah Zhu，Charlotte MUAdvisor:Shirley Yao,Wayne Luo 2 ABOUT THE AUTHERlogistics,capital flow and information flow.I

53、n addition,support will be provided in the form of vertical market-based industrial fund support,industry media matrix services,and cross-border service provider system docking.Drafters:Kinding Singapore Office(In alphabetical order)Melody Chen,Tate He,Aria Hu,Shuai Huang,Joy Huang,Zhenghao Jiang,Ma

54、tt Ma，Charlotte MU,Serena Shu,Nathan Xia,Junchao Zhu,Sarah ZhuAdvisors:Shirley Yao,Wayne Luo 3 ABOUT THE AUTHERThis document represents the authors personal views only and is for general information purposes only.It should not be used as a substitute for professional advice.Although the information

55、in this document has been obtained from sources that we believe to be reliable at the time of publication,given that AIlaws and regulations are changing rapidly and judicial practice will be handled based on the actual circumstances of each case,we do not make any explicit or implicit guarantee on t

56、he accuracy,reliability,timeliness,and completeness of the content,views,and suggestions of this document.This document is for corporate reference only and should not be relied upon.Any decision made solely based on the whole or part of this document and the consequences thereof shall be the sole re

57、sponsibility of the actor.Theorganisations and authors of this document expressly disclaims any responsibility.There will be no further notice of changes or updates to relevant laws and regulations.DISCLAIMER4 DISCLAIMERPREFACEWith the rapid development of generative artificial intelligence(generati

58、ve AI)technology,its applications have spread into all aspects of social life.From e-commerce,manufacturing and financial services to education and entertainment,every field is influenced by this cutting-edge technology.Generative AI has not only brought unprecedented innovation and efficiency impro

59、vements,but it has also sparked many discussions and challenges related to ethics,politics,and economics.In this context,the importance of legal compliance is particularly prominent.It should be noted that compliance requirements vary among different countries and regions,which requires our in-depth

60、 knowledge and strict compliance.Southeast Asia,as one of the worlds rapidly developing economic regions,is also witnessing robust growth in the AI industry.Southeast Asian countries have distinct regional characteristics in terms of politics,culture,and legal systems.Therefore,compliance issues are

61、 particularly important in the application and development of generative AI in this region.The legal frameworks,AI related regulations,ethical standards,and government regulatory policies of different countries all require special attention and adherence to ensure the organic integration of technolo

62、gical innovation and legal compliance.The Southeast Asian region is experiencing rapid economic growth,and each country is also demonstrating great potential in scientific and technological innovation.However,due to political,economic,and cultural differences,the legal regulations and compliance req

63、uirements for generative AI vary from country to country.This creates unique challenges and demands for the AI industry in Southeast Asia in terms of compliance.通過對東南亞各國成式AI產業及監管框架的深分析，本書旨在全解析東南亞各國在成式智能領域的法律合規要求，幫助企業和從業者更好地理解和應對這復雜多變的法律環境。我們將從各國具體的法律法規出發，結合實際案例和應場景，深探討東南亞AI業的合規特點和應對策略。希望通過本書的研究和分享，能

64、夠為AI從業者提供有價值的指導和參考，推動成式智能技術在東南亞地區的健康、可持續發展。Through an in-depth analysis of the generative AI industry and regulatory frameworks in Southeast Asia countries,this white paper aims to comprehensively interpret the legal compliance requirements in the generative AI field across Southeast Asia,so as to h

65、elp enterprises and practitioners better understand and cope with this complex and changing legal environment.Starting from the specific laws and regulations of each country,and combining real-world cases and application scenarios,we will delve into the compliance characteristics of the AI industry

66、in Southeast Asia as well as response strategies.Through the research and insights shared in this white paper,we hope to provide valuable guidance and references for AI practitioners,thereby promoting the healthy and sustainable development of generative AI technology in Southeast Asia.5 PREFACEappl

67、ication and development of generative AI in this region.The legal frameworks,AI related regulations,ethical standards,and government regulatory policies of different countries all require special attention and adherence to ensure the organic integration of technological innovation and legal complian

68、ce.The Southeast Asian region is experiencing rapid economic growth,and each country is also demonstrating great potential in scientific and technological innovation.However,due to political,economic,and cultural differences,the legal regulations and compliance requirements for generative AI vary fr

69、om country to country.This creates unique challenges and demands for the AI industry in Southeast Asia in terms of compliance.Through an in-depth analysis of the generative AI industry and regulatory frameworks in Southeast Asia countries,this white paper aims to comprehensively interpret the legal

70、compliance requirements in the generative AI field across Southeast Asia,so as to help enterprises and practitioners better understand and cope with this complex and changing legal environment.Starting from the specific laws and regulations of each country,and combining real-world cases and applicat

71、ion scenarios,we will delve into the compliance characteristics of the AI industry in Southeast Asia as well as response strategies.Through the research and insights shared in this white paper,we hope to provide valuable guidance and references for AI practitioners,thereby promoting the healthy and

72、sustainable development of generative AI technology in Southeast Asia.6 PREFACEI.SINGAPORE(I)Generative AI Industry and Regulation in Singapore1.Current status and position of generative AI in Singapore2.Regulatory framework for generative AI in Singapore(II)Core Issues of Generative AI Compliance i

73、n Singapore1.Relationship and characterization between foundation models and AI products2.Compliance of large model pre-training using public data3.Data localization and data cross-border4.Developer security responsibility5.Content security6.The copyrightability of intellectual property of generated

74、 content7.Transparency of large model8.User rights protection(I)Generative AI Industry and Regulation in Singapore1.Current status and position of generative AI in Singapore(1)Generative AI industry policies Singapore adopts a pectoral approach toward AI governance,that is,managing the use of AI thr

75、ough regulators of various industries.These regulators primarily govern through issuing non-binding guidelines and recommendations(soft law measures)rather than through mandatory regulations(hard law measures).Financial sector.Monetary Authority of Singapore(MAS),the central bank and integrated fina

76、ncial regulator of Singapore,is the first sectoral regulator to take action in AI governance.In 2018,MAS and the financial industry jointly created the FEAT principles(Fairness,Ethics,Accountability,and Transparency)to guide the responsible use of AI.In 2019,MAS announced that it was working with fi

77、nancial industry partners to create the Veritas framework to provide financial institutions with a verifiable way to incorporate the FEAT principles into their AI and data analytics-driven solutions.The FEAT principles and Veritas framework is part of Singapores national AI strategy,designed to help

78、 build a progressive and trusted environment for AI adoption within the financial sector.信息通信領域。信息通信媒體發展局（IMDA）和個數據保護委員會（PDPC）是智能治理中最活躍的監管機構。2019年以來，IMDA和PDPC每年都推出了與智能治理相關的指南或倡議。2019年，兩機構在達沃斯世界經濟論壇年會上發布了個智能治理模型框架。該框架旨在為私營部組織在部署智能解決案時提供可實施的指導，解決關鍵的倫理和治理問題。依據國際隱私專業員協會（IAPP）的相關報告，2020年，上述兩機構更新了該模型框架，發布

79、了第版，并推出了組織實施和我評估指南，幫助組織評估其智能治理實踐與模型框架的匹配程度，還發布了案例匯編，展示了組織如何實施負責任的智能治理實踐。Information and communications sector.The Info-communications Media Development Authority(IMDA)and Personal Data Protection Commission(PDPC)have been the most active regulators in the AI governance,launching AI governance-relate

80、d guidelines or initiatives every year since 2019.In 2019,IMDA and PDPC launched their first edition of Model AI Governance Framework at the annual meeting of the World Economic Forum in Davos.The framework is designed to provide private sector organizations with readily implementable guidance on ke

81、y ethical and governance issues when deploying AI solutions.Based on a related report by the International Association of Privacy Professionals(IAPP),IMDA and PDPC updated the Model Framework and issued a second edition in 2020.They also launched an Implementation and Self-Assessment Guide for Organ

82、izations,which aims to help organizations assess the alignment of their AI governance practices with the Model Framework,and a Compendium of Use Cases,which illustrates how organizations implement responsible AI governance practices.醫療衛領域。2021年10，新加坡衛部發布了醫療智能指南，提患者對智能在醫療中使的信任。該指南是與衛科學局和綜合健康信息系統（現稱為S

83、ynapxe）共同完成的，補充了新加坡對智能醫療設備監管的空。Healthcare sector.In October 2021,the Ministry of Health issued the AI in Healthcare Guidelines to improve the trust of the patients in the use of AI in healthcare.These guidelines were co-developed with the Health Sciences Authority(HSA)and the Integrated Health Infor

84、mation Systems(now known as Synapxe).They have complemented the gaps in Singapores regulation of AI medical devices.成式智能評估沙箱。新加坡信息通信媒體發展局在2024年2初與新加坡企業發展局合作啟動了成式智能評估沙箱。成式智能評估沙箱將為選中的中企業提供AIGC解決案，并根據中企業提供的反饋結果評估這些解決案，最終在本地中企業中推成式智能應。Generative AI Evaluation Sandbox.In early February,the IMDA of Singap

85、ore launched the Generative AI Evaluation Sandbox in cooperation with Enterprise Singapore.The Generative AI Evaluation Sandbox will provide AIGC solutions for selected small-and medium-sized enterprises(SMEs)and evaluate these solutions based on the SMEs feedback,ultimately promoting generative AI

86、applications across local SMEs.（2）成式AI企業案例(2)Generative AI company cases 活躍在新加坡的成式AI企業：Generative AI companies operating in Singapore:AI RudderAI RudderAI Rudder成于2019年，是家領先的對話式AI平臺和解決案提供商，專注于潛在客戶成、客戶服務和技術優化。其主要產品包括智能語助、AI聊天機器、VoiceGPT等。AI Rudder的技術在融業中表現尤為出，優化了動語識別（ASR）模型，提升了客戶服務的準確性和效率。AI Rudder的模

87、型持新加坡當地IP訪問。Founded in 2019,AI Rudder is a leading provider of conversational AI platforms and solutions focused on lead generation,customer service and technology optimization.Its main products include AI voice assistant,AI chatbot,VoiceGPT,etc.AI Rudders technology has performed particularly well

88、in the financial sector,optimizing the automatic speech recognition(ASR)model and improving the accuracy and efficiency of customer service.AI Rudders model supports access from local IP addresses in Singapore.ADVANCE.AIADVANCE.AI址：https:/www.advance.ai/Website:https:/www.advance.ai/ADVANCE.AI成于2016

89、年，總部位于新加坡，并在全球如中國、印尼、菲律賓、來亞等多地設有本地客戶持團隊，其公司是東南亞獨獸Advance Intelligence Group，集團業務覆蓋12個市場。該公司利AI、數據和云計算技術提供數字化解決案，主要服務包括臉識別、活體檢測、OCR技術持的數字身份驗證、數據與機器學習持的險管理和信貸評估、企業級數據分析等。Established in 2016,ADVANCE.AI is headquartered in Singapore and has local customer support teams in many places around the world,in

90、cluding China,Indonesia,Philippines,and Malaysia.Its parent company is Advance Intelligence Group,a Southeast Asia unicorn,and the groups business covers 12 markets.The company uses AI,big data,and cloud computing technologies to provide digital solutions,including face recognition,liveness detectio

91、n,digital identity verification supported by OCR technology,risk management and credit assessment supported by big data and machine learning,and enterprise-level data analysis.ADVANCE.AI 提供混合部署案，可根據企業實際情況選擇合適的式運應程序（本地或云端），持新加坡當地IP訪問。ADVANCE.AI 的模型本身不提供 Google Trends 功能。如果需要使 Google Trends 數據，通常需要通過調

92、 Google Trends API 來實現。ADVANCE.AI provides hybrid deployment solutions that allow users to choose the appropriate way to run applications(local or cloud)based on the actual situation of the enterprise,with support for local IP access in Singapore.ADVANCE.AIs big model itself does not provide Google

93、Trends functionality.If Google Trends data is required,it is typically achieved by calling the Google Trends API.科訊（iFLYTEK）iFLYTEK址：https:/ Co.,Ltd.,founded in 1999,is a listed intelligent voice and AI company.iFLYTEK provides products and services including intelligent voice,computer vision,natura

94、l language processing,and cognitive intelligence in China and worldwide.科訊將新加坡作為其海外業務的第站和戰略中。2024年5，科訊新加坡辦公室正式開業，并宣布將在新加坡建區域總部。為實現國際化戰略布局，科訊制定了“1+4”戰略，即個以新加坡為中的訊開放平臺國際站與四項戰略投資：本地化投資、技術投資、產品創新投資、合作伙伴計劃投資。iFLYTEK has made Singapore the first stop and strategic center for its overseas business.In May 2

95、024,the iFLYTEK Singapore office opened and announced the establishment of a regional headquarters in Singapore.To realize the international strategic layout,iFLYTEK has formulated the 1+4 strategy,consisting of one international station of the iFLYTEK Open Platform centered in Singapore and four st

96、rategic investments:localized investment,technology investment,product innovation investment,and partner program investment.科訊在新加坡推出了星認知模型，星認知模型是個基于云服務的模型平臺，具備跨領域的知識和語理解能，能夠基于然對話式理解與執任務。星認知模型持當地IP訪問，未限制其在特定國家或地區的訪問權限。iFLYTEK launched the Spark Cognitive Model in Singapore.It is a large model platfor

97、m based on cloud services,with cross-domain knowledge and language understanding capabilities,enabling it to understand and perform tasks based on a natural dialog approach.Spark Cognitive Model supports local IP access and does not restrict its access rights in specific countries or regions.星認知模型本身

98、不直接提供Google Trends功能，但可以通過API集成實現對Google Trends數據的處理和分析。The Spark Cognitive Model itself does not directly provide Google Trends functionality,but it can process and analyze Google Trends data through integrated API.字節跳動ByteDance字節跳動成于2012年，是家跨國互聯技術公司，主要產品包括抖、今頭條、視頻、書、剪映等。ByteDance,founded in 2012,i

99、s a multinational Internet technology company whose main products include TikTok,Toutiao,Watermelon Video,Feishu,CapCut,etc.字節跳動積極拓展新加坡市場，將TikTok業務總部遷新加坡，并在新加坡設了SPRING公司，專注于開發和推AI應。2023年，字節跳動在海外上線基于云雀語模型（現名為“包模型”）創建的AI具平臺“ChitChop”，為戶提供200余種作、活場景智能機器服務。ByteDance is actively expanding its market in S

100、ingapore.It has moved the TikTok business headquarters to Singapore,and set up SPRING company in Singapore to focus on the development and promotion of AI applications.In 2023,ByteDance launched ChitChop overseas,an AI tool platform created based on the Skylark Large Language Model(now known as Doub

101、ao Model),providing users with over 200 smart robot services for work and life scenarios.字節跳動推出的包模型持新加坡當地IP訪問。The Doubao Model launched by ByteDance supports access from local IP addresses in Singapore.其他國家投資該國情況：Foreign investments in Singapore新加坡已然成為東南亞地區最受歡迎的初創企業投資的地，尤其在智能領域。新加坡的智能獨獸公司和智能初創企業吸引了來

102、全球各地的投資者。Singapore has become the most popular destination for startup investment in Southeast Asia,especially in the AI field.Singapores AI unicorns and AI startups have attracted investors from around the world.美國投資者積極參與新加坡AI企業的融資活動。U.S.investors have been active in fundraising activities for AI c

103、ompanies in Singapore.?根據新加坡經濟報，新加坡已與美國晶制造商英偉達（Nvidia）達成共識，由英偉達公司幫助、持和參與新加坡NAIS 2.0，深化雙在AI技術領域的合作，共同創建個持11種語的語模型，以發展新加坡本AI技術，并將在新加坡新建臺超級計算機。?According to Singapores Economic Times，Singapore has reached a consensus with U.S.chip maker Nvidia that Nvidia will help,support and participate in Singapores

104、 NAIS 2.0,to deepen their cooperation in AI technology,to co-create a large language model supporting 11 languages to develop Singapores homegrown AI technology,and to build a new supercomputer in Singapore.?在新加坡舉的第屆AWS東盟峰會上，亞遜云科技（Amazon Web Services，簡稱AWS）宣布計劃從2024年到2028年額外投資120億新元（約合88.8 億美元）于擴展其在

105、新加坡現有的云基礎設施，以滿該國對云技術和服務益增的客戶需求。AWS還宣布了項名為AWS AI Spring的新計劃，以加速AI（包括成式智能）在新加坡的采。重點將是為公共部、當地企業、初創企業、社區、研發組織和勞動開發這些技術。前AWS已就該計劃與新加坡資媒局IMDA簽署了份意向備忘錄。?At the 10th AWS ASEAN Summit held in Singapore,Amazon Web Services(AWS)announced plans to invest an additional SGD12 billion(approximately USD8.88 billion

106、)from 2024 to 2028 to expand its existing cloud infrastructure in Singapore to meet the countrys growing customer demand for cloud technology and services.AWS also announced a new initiative called AWS AI Spring to accelerate the adoption of AI,including generative AI,in Singapore.The focus will be

107、on developing these technologies for the public sectors,local businesses,start-ups,communities,R&D organizations and the workforce.Currently,AWS has signed a memorandum of intent with Singapores Infocomm Media Development Authority(IMDA)for the initiative.其他國家如班、印度尼亞和來亞也在新加坡AI領域有定的投資活動。例如，Bolttech公司

108、還從班和新加坡EDBI獲得了3000萬美元的投資。此外，新加坡也吸引了來東南亞其他國家如泰國和越南的企業前來設研發中或進技術合作。Other countries such as Spain,Indonesia and Malaysia also have some investment activities in the AI field in Singapore.Bolttech,for example,also received USD30 million in investments from EDBI in both Spain and Singapore.In addition,Si

109、ngapore has also attracted enterprises from other Southeast Asian countries such as Thailand and Vietnam to set up R&D centers or conduct technical cooperation.2 I.Singapore progressive and trusted environment for AI adoption within the financial sector.Information and communications sector.The Info

110、-communications Media Development Authority(IMDA)and Personal Data Protection Commission(PDPC)have been the most active regulators in the AI governance,launching AI governance-related guidelines or initiatives every year since 2019.In 2019,IMDA and PDPC launched their first edition of Model AI Gover

111、nance Framework at the annual meeting of the World Economic Forum in Davos.The framework is designed to provide private sector organizations with readily implementable guidance on key ethical and governance issues when deploying AI solutions.Based on a related report by the International Association

112、 of Privacy Professionals(IAPP),IMDA and PDPC updated the Model Framework and issued a second edition in 2020.They also launched an Implementation and Self-Assessment Guide for Organizations,which aims to help organizations assess the alignment of their AI governance practices with the Model Framewo

113、rk,and a Compendium of Use Cases,which illustrates how organizations implement responsible AI governance practices.Healthcare sector.In October 2021,the Ministry of Health issued the AI in Healthcare Guidelines to improve the trust of the patients in the use of AI in healthcare.These guidelines were

114、 co-developed with the Health Sciences Authority(HSA)and the Integrated Health Information Systems(now known as Synapxe).They have complemented the gaps in Singapores regulation of AI medical devices.成式智能評估沙箱。新加坡信息通信媒體發展局在2024年2初與新加坡企業發展局合作啟動了成式智能評估沙箱。成式智能評估沙箱將為選中的中企業提供AIGC解決案，并根據中企業提供的反饋結果評估這些解決案，最

115、終在本地中企業中推成式智能應。Generative AI Evaluation Sandbox.In early February,the IMDA of Singapore launched the Generative AI Evaluation Sandbox in cooperation with Enterprise Singapore.The Generative AI Evaluation Sandbox will provide AIGC solutions for selected small-and medium-sized enterprises(SMEs)and eva

116、luate these solutions based on the SMEs feedback,ultimately promoting generative AI applications across local SMEs.（2）成式AI企業案例(2)Generative AI company cases 活躍在新加坡的成式AI企業：Generative AI companies operating in Singapore:AI RudderAI RudderAI Rudder成于2019年，是家領先的對話式AI平臺和解決案提供商，專注于潛在客戶成、客戶服務和技術優化。其主要產品包括智

117、能語助、AI聊天機器、VoiceGPT等。AI Rudder的技術在融業中表現尤為出，優化了動語識別（ASR）模型，提升了客戶服務的準確性和效率。AI Rudder的模型持新加坡當地IP訪問。Founded in 2019,AI Rudder is a leading provider of conversational AI platforms and solutions focused on lead generation,customer service and technology optimization.Its main products include AI voice assi

118、stant,AI chatbot,VoiceGPT,etc.AI Rudders technology has performed particularly well in the financial sector,optimizing the automatic speech recognition(ASR)model and improving the accuracy and efficiency of customer service.AI Rudders model supports access from local IP addresses in Singapore.ADVANC

119、E.AIADVANCE.AI址：https:/www.advance.ai/Website:https:/www.advance.ai/ADVANCE.AI成于2016年，總部位于新加坡，并在全球如中國、印尼、菲律賓、來亞等多地設有本地客戶持團隊，其公司是東南亞獨獸Advance Intelligence Group，集團業務覆蓋12個市場。該公司利AI、數據和云計算技術提供數字化解決案，主要服務包括臉識別、活體檢測、OCR技術持的數字身份驗證、數據與機器學習持的險管理和信貸評估、企業級數據分析等。Established in 2016,ADVANCE.AI is headquartered

120、in Singapore and has local customer support teams in many places around the world,including China,Indonesia,Philippines,and Malaysia.Its parent company is Advance Intelligence Group,a Southeast Asia unicorn,and the groups business covers 12 markets.The company uses AI,big data,and cloud computing te

121、chnologies to provide digital solutions,including face recognition,liveness detection,digital identity verification supported by OCR technology,risk management and credit assessment supported by big data and machine learning,and enterprise-level data analysis.ADVANCE.AI 提供混合部署案，可根據企業實際情況選擇合適的式運應程序（本

122、地或云端），持新加坡當地IP訪問。ADVANCE.AI 的模型本身不提供 Google Trends 功能。如果需要使 Google Trends 數據，通常需要通過調 Google Trends API 來實現。ADVANCE.AI provides hybrid deployment solutions that allow users to choose the appropriate way to run applications(local or cloud)based on the actual situation of the enterprise,with support fo

123、r local IP access in Singapore.ADVANCE.AIs big model itself does not provide Google Trends functionality.If Google Trends data is required,it is typically achieved by calling the Google Trends API.科訊（iFLYTEK）iFLYTEK址：https:/ Co.,Ltd.,founded in 1999,is a listed intelligent voice and AI company.iFLYT

124、EK provides products and services including intelligent voice,computer vision,natural language processing,and cognitive intelligence in China and worldwide.科訊將新加坡作為其海外業務的第站和戰略中。2024年5，科訊新加坡辦公室正式開業，并宣布將在新加坡建區域總部。為實現國際化戰略布局，科訊制定了“1+4”戰略，即個以新加坡為中的訊開放平臺國際站與四項戰略投資：本地化投資、技術投資、產品創新投資、合作伙伴計劃投資。iFLYTEK has m

125、ade Singapore the first stop and strategic center for its overseas business.In May 2024,the iFLYTEK Singapore office opened and announced the establishment of a regional headquarters in Singapore.To realize the international strategic layout,iFLYTEK has formulated the 1+4 strategy,consisting of one

126、international station of the iFLYTEK Open Platform centered in Singapore and four strategic investments:localized investment,technology investment,product innovation investment,and partner program investment.科訊在新加坡推出了星認知模型，星認知模型是個基于云服務的模型平臺，具備跨領域的知識和語理解能，能夠基于然對話式理解與執任務。星認知模型持當地IP訪問，未限制其在特定國家或地區的訪問權限

127、。iFLYTEK launched the Spark Cognitive Model in Singapore.It is a large model platform based on cloud services,with cross-domain knowledge and language understanding capabilities,enabling it to understand and perform tasks based on a natural dialog approach.Spark Cognitive Model supports local IP acc

128、ess and does not restrict its access rights in specific countries or regions.星認知模型本身不直接提供Google Trends功能，但可以通過API集成實現對Google Trends數據的處理和分析。The Spark Cognitive Model itself does not directly provide Google Trends functionality,but it can process and analyze Google Trends data through integrated API.

129、字節跳動ByteDance字節跳動成于2012年，是家跨國互聯技術公司，主要產品包括抖、今頭條、視頻、書、剪映等。ByteDance,founded in 2012,is a multinational Internet technology company whose main products include TikTok,Toutiao,Watermelon Video,Feishu,CapCut,etc.字節跳動積極拓展新加坡市場，將TikTok業務總部遷新加坡，并在新加坡設了SPRING公司，專注于開發和推AI應。2023年，字節跳動在海外上線基于云雀語模型（現名為“包模型”）創建的

130、AI具平臺“ChitChop”，為戶提供200余種作、活場景智能機器服務。ByteDance is actively expanding its market in Singapore.It has moved the TikTok business headquarters to Singapore,and set up SPRING company in Singapore to focus on the development and promotion of AI applications.In 2023,ByteDance launched ChitChop overseas,an

131、AI tool platform created based on the Skylark Large Language Model(now known as Doubao Model),providing users with over 200 smart robot services for work and life scenarios.字節跳動推出的包模型持新加坡當地IP訪問。The Doubao Model launched by ByteDance supports access from local IP addresses in Singapore.其他國家投資該國情況：For

132、eign investments in Singapore新加坡已然成為東南亞地區最受歡迎的初創企業投資的地，尤其在智能領域。新加坡的智能獨獸公司和智能初創企業吸引了來全球各地的投資者。Singapore has become the most popular destination for startup investment in Southeast Asia,especially in the AI field.Singapores AI unicorns and AI startups have attracted investors from around the world.美國投

133、資者積極參與新加坡AI企業的融資活動。U.S.investors have been active in fundraising activities for AI companies in Singapore.?根據新加坡經濟報，新加坡已與美國晶制造商英偉達（Nvidia）達成共識，由英偉達公司幫助、持和參與新加坡NAIS 2.0，深化雙在AI技術領域的合作，共同創建個持11種語的語模型，以發展新加坡本AI技術，并將在新加坡新建臺超級計算機。?According to Singapores Economic Times，Singapore has reached a consensus wi

134、th U.S.chip maker Nvidia that Nvidia will help,support and participate in Singapores NAIS 2.0,to deepen their cooperation in AI technology,to co-create a large language model supporting 11 languages to develop Singapores homegrown AI technology,and to build a new supercomputer in Singapore.?在新加坡舉的第屆

135、AWS東盟峰會上，亞遜云科技（Amazon Web Services，簡稱AWS）宣布計劃從2024年到2028年額外投資120億新元（約合88.8 億美元）于擴展其在新加坡現有的云基礎設施，以滿該國對云技術和服務益增的客戶需求。AWS還宣布了項名為AWS AI Spring的新計劃，以加速AI（包括成式智能）在新加坡的采。重點將是為公共部、當地企業、初創企業、社區、研發組織和勞動開發這些技術。前AWS已就該計劃與新加坡資媒局IMDA簽署了份意向備忘錄。?At the 10th AWS ASEAN Summit held in Singapore,Amazon Web Services(AWS

136、)announced plans to invest an additional SGD12 billion(approximately USD8.88 billion)from 2024 to 2028 to expand its existing cloud infrastructure in Singapore to meet the countrys growing customer demand for cloud technology and services.AWS also announced a new initiative called AWS AI Spring to a

137、ccelerate the adoption of AI,including generative AI,in Singapore.The focus will be on developing these technologies for the public sectors,local businesses,start-ups,communities,R&D organizations and the workforce.Currently,AWS has signed a memorandum of intent with Singapores Infocomm Media Develo

138、pment Authority(IMDA)for the initiative.其他國家如班、印度尼亞和來亞也在新加坡AI領域有定的投資活動。例如，Bolttech公司還從班和新加坡EDBI獲得了3000萬美元的投資。此外，新加坡也吸引了來東南亞其他國家如泰國和越南的企業前來設研發中或進技術合作。Other countries such as Spain,Indonesia and Malaysia also have some investment activities in the AI field in Singapore.Bolttech,for example,also receiv

139、ed USD30 million in investments from EDBI in both Spain and Singapore.In addition,Singapore has also attracted enterprises from other Southeast Asian countries such as Thailand and Vietnam to set up R&D centers or conduct technical cooperation.3 I.Singaporethe Health Sciences Authority(HSA)and the I

140、ntegrated Health Information Systems(now known as Synapxe).They have complemented the gaps in Singapores regulation of AI medical devices.Generative AI Evaluation Sandbox.In early February,the IMDA of Singapore launched the Generative AI Evaluation Sandbox in cooperation with Enterprise Singapore.Th

141、e Generative AI Evaluation Sandbox will provide AIGC solutions for selected small-and medium-sized enterprises(SMEs)and evaluate these solutions based on the SMEs feedback,ultimately promoting generative AI applications across local SMEs.(2)Generative AI company cases Generative AI companies operati

142、ng in Singapore:AI RudderFounded in 2019,AI Rudder is a leading provider of conversational AI platforms and solutions focused on lead generation,customer service and technology optimization.Its main products include AI voice assistant,AI chatbot,VoiceGPT,etc.AI Rudders technology has performed parti

143、cularly well in the financial sector,optimizing the automatic speech recognition(ASR)model and improving the accuracy and efficiency of customer service.AI Rudders model supports access from local IP addresses in Singapore.ADVANCE.AIADVANCE.AI址：https:/www.advance.ai/Website:https:/www.advance.ai/ADV

144、ANCE.AI成于2016年，總部位于新加坡，并在全球如中國、印尼、菲律賓、來亞等多地設有本地客戶持團隊，其公司是東南亞獨獸Advance Intelligence Group，集團業務覆蓋12個市場。該公司利AI、數據和云計算技術提供數字化解決案，主要服務包括臉識別、活體檢測、OCR技術持的數字身份驗證、數據與機器學習持的險管理和信貸評估、企業級數據分析等。Established in 2016,ADVANCE.AI is headquartered in Singapore and has local customer support teams in many places around

145、 the world,including China,Indonesia,Philippines,and Malaysia.Its parent company is Advance Intelligence Group,a Southeast Asia unicorn,and the groups business covers 12 markets.The company uses AI,big data,and cloud computing technologies to provide digital solutions,including face recognition,live

146、ness detection,digital identity verification supported by OCR technology,risk management and credit assessment supported by big data and machine learning,and enterprise-level data analysis.ADVANCE.AI 提供混合部署案，可根據企業實際情況選擇合適的式運應程序（本地或云端），持新加坡當地IP訪問。ADVANCE.AI 的模型本身不提供 Google Trends 功能。如果需要使 Google Tren

147、ds 數據，通常需要通過調 Google Trends API 來實現。ADVANCE.AI provides hybrid deployment solutions that allow users to choose the appropriate way to run applications(local or cloud)based on the actual situation of the enterprise,with support for local IP access in Singapore.ADVANCE.AIs big model itself does not pr

148、ovide Google Trends functionality.If Google Trends data is required,it is typically achieved by calling the Google Trends API.科訊（iFLYTEK）iFLYTEK址：https:/ Co.,Ltd.,founded in 1999,is a listed intelligent voice and AI company.iFLYTEK provides products and services including intelligent voice,computer

149、vision,natural language processing,and cognitive intelligence in China and worldwide.科訊將新加坡作為其海外業務的第站和戰略中。2024年5，科訊新加坡辦公室正式開業，并宣布將在新加坡建區域總部。為實現國際化戰略布局，科訊制定了“1+4”戰略，即個以新加坡為中的訊開放平臺國際站與四項戰略投資：本地化投資、技術投資、產品創新投資、合作伙伴計劃投資。iFLYTEK has made Singapore the first stop and strategic center for its overseas busi

150、ness.In May 2024,the iFLYTEK Singapore office opened and announced the establishment of a regional headquarters in Singapore.To realize the international strategic layout,iFLYTEK has formulated the 1+4 strategy,consisting of one international station of the iFLYTEK Open Platform centered in Singapor

151、e and four strategic investments:localized investment,technology investment,product innovation investment,and partner program investment.科訊在新加坡推出了星認知模型，星認知模型是個基于云服務的模型平臺，具備跨領域的知識和語理解能，能夠基于然對話式理解與執任務。星認知模型持當地IP訪問，未限制其在特定國家或地區的訪問權限。iFLYTEK launched the Spark Cognitive Model in Singapore.It is a large

152、model platform based on cloud services,with cross-domain knowledge and language understanding capabilities,enabling it to understand and perform tasks based on a natural dialog approach.Spark Cognitive Model supports local IP access and does not restrict its access rights in specific countries or re

153、gions.星認知模型本身不直接提供Google Trends功能，但可以通過API集成實現對Google Trends數據的處理和分析。The Spark Cognitive Model itself does not directly provide Google Trends functionality,but it can process and analyze Google Trends data through integrated API.字節跳動ByteDance字節跳動成于2012年，是家跨國互聯技術公司，主要產品包括抖、今頭條、視頻、書、剪映等。ByteDance,foun

154、ded in 2012,is a multinational Internet technology company whose main products include TikTok,Toutiao,Watermelon Video,Feishu,CapCut,etc.字節跳動積極拓展新加坡市場，將TikTok業務總部遷新加坡，并在新加坡設了SPRING公司，專注于開發和推AI應。2023年，字節跳動在海外上線基于云雀語模型（現名為“包模型”）創建的AI具平臺“ChitChop”，為戶提供200余種作、活場景智能機器服務。ByteDance is actively expanding it

155、s market in Singapore.It has moved the TikTok business headquarters to Singapore,and set up SPRING company in Singapore to focus on the development and promotion of AI applications.In 2023,ByteDance launched ChitChop overseas,an AI tool platform created based on the Skylark Large Language Model(now

156、known as Doubao Model),providing users with over 200 smart robot services for work and life scenarios.字節跳動推出的包模型持新加坡當地IP訪問。The Doubao Model launched by ByteDance supports access from local IP addresses in Singapore.其他國家投資該國情況：Foreign investments in Singapore新加坡已然成為東南亞地區最受歡迎的初創企業投資的地，尤其在智能領域。新加坡的智能獨獸

157、公司和智能初創企業吸引了來全球各地的投資者。Singapore has become the most popular destination for startup investment in Southeast Asia,especially in the AI field.Singapores AI unicorns and AI startups have attracted investors from around the world.美國投資者積極參與新加坡AI企業的融資活動。U.S.investors have been active in fundraising activi

158、ties for AI companies in Singapore.?根據新加坡經濟報，新加坡已與美國晶制造商英偉達（Nvidia）達成共識，由英偉達公司幫助、持和參與新加坡NAIS 2.0，深化雙在AI技術領域的合作，共同創建個持11種語的語模型，以發展新加坡本AI技術，并將在新加坡新建臺超級計算機。?According to Singapores Economic Times，Singapore has reached a consensus with U.S.chip maker Nvidia that Nvidia will help,support and participate

159、in Singapores NAIS 2.0,to deepen their cooperation in AI technology,to co-create a large language model supporting 11 languages to develop Singapores homegrown AI technology,and to build a new supercomputer in Singapore.?在新加坡舉的第屆AWS東盟峰會上，亞遜云科技（Amazon Web Services，簡稱AWS）宣布計劃從2024年到2028年額外投資120億新元（約合8

160、8.8 億美元）于擴展其在新加坡現有的云基礎設施，以滿該國對云技術和服務益增的客戶需求。AWS還宣布了項名為AWS AI Spring的新計劃，以加速AI（包括成式智能）在新加坡的采。重點將是為公共部、當地企業、初創企業、社區、研發組織和勞動開發這些技術。前AWS已就該計劃與新加坡資媒局IMDA簽署了份意向備忘錄。?At the 10th AWS ASEAN Summit held in Singapore,Amazon Web Services(AWS)announced plans to invest an additional SGD12 billion(approximately US

161、D8.88 billion)from 2024 to 2028 to expand its existing cloud infrastructure in Singapore to meet the countrys growing customer demand for cloud technology and services.AWS also announced a new initiative called AWS AI Spring to accelerate the adoption of AI,including generative AI,in Singapore.The f

162、ocus will be on developing these technologies for the public sectors,local businesses,start-ups,communities,R&D organizations and the workforce.Currently,AWS has signed a memorandum of intent with Singapores Infocomm Media Development Authority(IMDA)for the initiative.其他國家如班、印度尼亞和來亞也在新加坡AI領域有定的投資活動。

163、例如，Bolttech公司還從班和新加坡EDBI獲得了3000萬美元的投資。此外，新加坡也吸引了來東南亞其他國家如泰國和越南的企業前來設研發中或進技術合作。Other countries such as Spain,Indonesia and Malaysia also have some investment activities in the AI field in Singapore.Bolttech,for example,also received USD30 million in investments from EDBI in both Spain and Singapore.I

164、n addition,Singapore has also attracted enterprises from other Southeast Asian countries such as Thailand and Vietnam to set up R&D centers or conduct technical cooperation.4 I.Singapore Singapore.ADVANCE.AIWebsite:https:/www.advance.ai/Established in 2016,ADVANCE.AI is headquartered in Singapore an

165、d has local customer support teams in many places around the world,including China,Indonesia,Philippines,and Malaysia.Its parent company is Advance Intelligence Group,a Southeast Asia unicorn,and the groups business covers 12 markets.The company uses AI,big data,and cloud computing technologies to p

166、rovide digital solutions,including face recognition,liveness detection,digital identity verification supported by OCR technology,risk management and credit assessment supported by big data and machine learning,and enterprise-level data analysis.ADVANCE.AI provides hybrid deployment solutions that al

167、low users to choose the appropriate way to run applications(local or cloud)based on the actual situation of the enterprise,with support for local IP access in Singapore.ADVANCE.AIs big model itself does not provide Google Trends functionality.If Google Trends data is required,it is typically achieve

168、d by calling the Google Trends API.iFLYTEKWebsite:https:/ Co.,Ltd.,founded in 1999,is a listed intelligent voice and AI company.iFLYTEK provides products and services including intelligent voice,computer vision,natural language processing,and cognitive intelligence in China and worldwide.科訊將新加坡作為其海外

169、業務的第站和戰略中。2024年5，科訊新加坡辦公室正式開業，并宣布將在新加坡建區域總部。為實現國際化戰略布局，科訊制定了“1+4”戰略，即個以新加坡為中的訊開放平臺國際站與四項戰略投資：本地化投資、技術投資、產品創新投資、合作伙伴計劃投資。iFLYTEK has made Singapore the first stop and strategic center for its overseas business.In May 2024,the iFLYTEK Singapore office opened and announced the establishment of a region

170、al headquarters in Singapore.To realize the international strategic layout,iFLYTEK has formulated the 1+4 strategy,consisting of one international station of the iFLYTEK Open Platform centered in Singapore and four strategic investments:localized investment,technology investment,product innovation i

171、nvestment,and partner program investment.科訊在新加坡推出了星認知模型，星認知模型是個基于云服務的模型平臺，具備跨領域的知識和語理解能，能夠基于然對話式理解與執任務。星認知模型持當地IP訪問，未限制其在特定國家或地區的訪問權限。iFLYTEK launched the Spark Cognitive Model in Singapore.It is a large model platform based on cloud services,with cross-domain knowledge and language understanding ca

172、pabilities,enabling it to understand and perform tasks based on a natural dialog approach.Spark Cognitive Model supports local IP access and does not restrict its access rights in specific countries or regions.星認知模型本身不直接提供Google Trends功能，但可以通過API集成實現對Google Trends數據的處理和分析。The Spark Cognitive Model i

173、tself does not directly provide Google Trends functionality,but it can process and analyze Google Trends data through integrated API.字節跳動ByteDance字節跳動成于2012年，是家跨國互聯技術公司，主要產品包括抖、今頭條、視頻、書、剪映等。ByteDance,founded in 2012,is a multinational Internet technology company whose main products include TikTok,To

174、utiao,Watermelon Video,Feishu,CapCut,etc.字節跳動積極拓展新加坡市場，將TikTok業務總部遷新加坡，并在新加坡設了SPRING公司，專注于開發和推AI應。2023年，字節跳動在海外上線基于云雀語模型（現名為“包模型”）創建的AI具平臺“ChitChop”，為戶提供200余種作、活場景智能機器服務。ByteDance is actively expanding its market in Singapore.It has moved the TikTok business headquarters to Singapore,and set up SPRI

175、NG company in Singapore to focus on the development and promotion of AI applications.In 2023,ByteDance launched ChitChop overseas,an AI tool platform created based on the Skylark Large Language Model(now known as Doubao Model),providing users with over 200 smart robot services for work and life scen

176、arios.字節跳動推出的包模型持新加坡當地IP訪問。The Doubao Model launched by ByteDance supports access from local IP addresses in Singapore.其他國家投資該國情況：Foreign investments in Singapore新加坡已然成為東南亞地區最受歡迎的初創企業投資的地，尤其在智能領域。新加坡的智能獨獸公司和智能初創企業吸引了來全球各地的投資者。Singapore has become the most popular destination for startup investment i

177、n Southeast Asia,especially in the AI field.Singapores AI unicorns and AI startups have attracted investors from around the world.美國投資者積極參與新加坡AI企業的融資活動。U.S.investors have been active in fundraising activities for AI companies in Singapore.?根據新加坡經濟報，新加坡已與美國晶制造商英偉達（Nvidia）達成共識，由英偉達公司幫助、持和參與新加坡NAIS 2.0

178、，深化雙在AI技術領域的合作，共同創建個持11種語的語模型，以發展新加坡本AI技術，并將在新加坡新建臺超級計算機。?According to Singapores Economic Times，Singapore has reached a consensus with U.S.chip maker Nvidia that Nvidia will help,support and participate in Singapores NAIS 2.0,to deepen their cooperation in AI technology,to co-create a large languag

179、e model supporting 11 languages to develop Singapores homegrown AI technology,and to build a new supercomputer in Singapore.?在新加坡舉的第屆AWS東盟峰會上，亞遜云科技（Amazon Web Services，簡稱AWS）宣布計劃從2024年到2028年額外投資120億新元（約合88.8 億美元）于擴展其在新加坡現有的云基礎設施，以滿該國對云技術和服務益增的客戶需求。AWS還宣布了項名為AWS AI Spring的新計劃，以加速AI（包括成式智能）在新加坡的采。重點將是

180、為公共部、當地企業、初創企業、社區、研發組織和勞動開發這些技術。前AWS已就該計劃與新加坡資媒局IMDA簽署了份意向備忘錄。?At the 10th AWS ASEAN Summit held in Singapore,Amazon Web Services(AWS)announced plans to invest an additional SGD12 billion(approximately USD8.88 billion)from 2024 to 2028 to expand its existing cloud infrastructure in Singapore to meet

181、 the countrys growing customer demand for cloud technology and services.AWS also announced a new initiative called AWS AI Spring to accelerate the adoption of AI,including generative AI,in Singapore.The focus will be on developing these technologies for the public sectors,local businesses,start-ups,

182、communities,R&D organizations and the workforce.Currently,AWS has signed a memorandum of intent with Singapores Infocomm Media Development Authority(IMDA)for the initiative.其他國家如班、印度尼亞和來亞也在新加坡AI領域有定的投資活動。例如，Bolttech公司還從班和新加坡EDBI獲得了3000萬美元的投資。此外，新加坡也吸引了來東南亞其他國家如泰國和越南的企業前來設研發中或進技術合作。Other countries su

183、ch as Spain,Indonesia and Malaysia also have some investment activities in the AI field in Singapore.Bolttech,for example,also received USD30 million in investments from EDBI in both Spain and Singapore.In addition,Singapore has also attracted enterprises from other Southeast Asian countries such as

184、 Thailand and Vietnam to set up R&D centers or conduct technical cooperation.5 I.Singaporeand AI company.iFLYTEK provides products and services including intelligent voice,computer vision,natural language processing,and cognitive intelligence in China and worldwide.iFLYTEK has made Singapore the fir

185、st stop and strategic center for its overseas business.In May 2024,the iFLYTEK Singapore office opened and announced the establishment of a regional headquarters in Singapore.To realize the international strategic layout,iFLYTEK has formulated the 1+4 strategy,consisting of one international station

186、 of the iFLYTEK Open Platform centered in Singapore and four strategic investments:localized investment,technology investment,product innovation investment,and partner program investment.iFLYTEK launched the Spark Cognitive Model in Singapore.It is a large model platform based on cloud services,with

187、 cross-domain knowledge and language understanding capabilities,enabling it to understand and perform tasks based on a natural dialog approach.Spark Cognitive Model supports local IP access and does not restrict its access rights in specific countries or regions.The Spark Cognitive Model itself does

188、 not directly provide Google Trends functionality,but it can process and analyze Google Trends data through integrated API.ByteDanceByteDance,founded in 2012,is a multinational Internet technology company whose main products include TikTok,Toutiao,Watermelon Video,Feishu,CapCut,etc.字節跳動積極拓展新加坡市場，將Ti

189、kTok業務總部遷新加坡，并在新加坡設了SPRING公司，專注于開發和推AI應。2023年，字節跳動在海外上線基于云雀語模型（現名為“包模型”）創建的AI具平臺“ChitChop”，為戶提供200余種作、活場景智能機器服務。ByteDance is actively expanding its market in Singapore.It has moved the TikTok business headquarters to Singapore,and set up SPRING company in Singapore to focus on the development and pr

190、omotion of AI applications.In 2023,ByteDance launched ChitChop overseas,an AI tool platform created based on the Skylark Large Language Model(now known as Doubao Model),providing users with over 200 smart robot services for work and life scenarios.字節跳動推出的包模型持新加坡當地IP訪問。The Doubao Model launched by By

191、teDance supports access from local IP addresses in Singapore.其他國家投資該國情況：Foreign investments in Singapore新加坡已然成為東南亞地區最受歡迎的初創企業投資的地，尤其在智能領域。新加坡的智能獨獸公司和智能初創企業吸引了來全球各地的投資者。Singapore has become the most popular destination for startup investment in Southeast Asia,especially in the AI field.Singapores AI

192、unicorns and AI startups have attracted investors from around the world.美國投資者積極參與新加坡AI企業的融資活動。U.S.investors have been active in fundraising activities for AI companies in Singapore.?根據新加坡經濟報，新加坡已與美國晶制造商英偉達（Nvidia）達成共識，由英偉達公司幫助、持和參與新加坡NAIS 2.0，深化雙在AI技術領域的合作，共同創建個持11種語的語模型，以發展新加坡本AI技術，并將在新加坡新建臺超級計算機。?

193、According to Singapores Economic Times，Singapore has reached a consensus with U.S.chip maker Nvidia that Nvidia will help,support and participate in Singapores NAIS 2.0,to deepen their cooperation in AI technology,to co-create a large language model supporting 11 languages to develop Singapores home

194、grown AI technology,and to build a new supercomputer in Singapore.?在新加坡舉的第屆AWS東盟峰會上，亞遜云科技（Amazon Web Services，簡稱AWS）宣布計劃從2024年到2028年額外投資120億新元（約合88.8 億美元）于擴展其在新加坡現有的云基礎設施，以滿該國對云技術和服務益增的客戶需求。AWS還宣布了項名為AWS AI Spring的新計劃，以加速AI（包括成式智能）在新加坡的采。重點將是為公共部、當地企業、初創企業、社區、研發組織和勞動開發這些技術。前AWS已就該計劃與新加坡資媒局IMDA簽署了份意向

195、備忘錄。?At the 10th AWS ASEAN Summit held in Singapore,Amazon Web Services(AWS)announced plans to invest an additional SGD12 billion(approximately USD8.88 billion)from 2024 to 2028 to expand its existing cloud infrastructure in Singapore to meet the countrys growing customer demand for cloud technology

196、 and services.AWS also announced a new initiative called AWS AI Spring to accelerate the adoption of AI,including generative AI,in Singapore.The focus will be on developing these technologies for the public sectors,local businesses,start-ups,communities,R&D organizations and the workforce.Currently,

197、AWS has signed a memorandum of intent with Singapores Infocomm Media Development Authority(IMDA)for the initiative.其他國家如班、印度尼亞和來亞也在新加坡AI領域有定的投資活動。例如，Bolttech公司還從班和新加坡EDBI獲得了3000萬美元的投資。此外，新加坡也吸引了來東南亞其他國家如泰國和越南的企業前來設研發中或進技術合作。Other countries such as Spain,Indonesia and Malaysia also have some investme

198、nt activities in the AI field in Singapore.Bolttech,for example,also received USD30 million in investments from EDBI in both Spain and Singapore.In addition,Singapore has also attracted enterprises from other Southeast Asian countries such as Thailand and Vietnam to set up R&D centers or conduct tec

199、hnical cooperation.6 I.Singapore technology company whose main products include TikTok,Toutiao,Watermelon Video,Feishu,CapCut,etc.ByteDance is actively expanding its market in Singapore.It has moved the TikTok business headquarters to Singapore,and set up SPRING company in Singapore to focus on the

200、development and promotion of AI applications.In 2023,ByteDance launched ChitChop overseas,an AI tool platform created based on the Skylark Large Language Model(now known as Doubao Model),providing users with over 200 smart robot services for work and life scenarios.The Doubao Model launched by ByteD

201、ance supports access from local IP addresses in Singapore.Foreign investments in SingaporeSingapore has become the most popular destination for startup investment in Southeast Asia,especially in the AI field.Singapores AI unicorns and AI startups have attracted investors from around the world.U.S.in

202、vestors have been active in fundraising activities for AI companies in Singapore.According to Singapores Economic Times，Singapore has reached a consensus with U.S.chip maker Nvidia that Nvidia will help,support and participate in Singapores NAIS 2.0,to deepen their cooperation in AI technology,to co

203、-create a large language model supporting 11 languages to develop Singapores homegrown AI technology,and to build a new supercomputer in Singapore.?在新加坡舉的第屆AWS東盟峰會上，亞遜云科技（Amazon Web Services，簡稱AWS）宣布計劃從2024年到2028年額外投資120億新元（約合88.8 億美元）于擴展其在新加坡現有的云基礎設施，以滿該國對云技術和服務益增的客戶需求。AWS還宣布了項名為AWS AI Spring的新計劃，以

204、加速AI（包括成式智能）在新加坡的采。重點將是為公共部、當地企業、初創企業、社區、研發組織和勞動開發這些技術。前AWS已就該計劃與新加坡資媒局IMDA簽署了份意向備忘錄。?At the 10th AWS ASEAN Summit held in Singapore,Amazon Web Services(AWS)announced plans to invest an additional SGD12 billion(approximately USD8.88 billion)from 2024 to 2028 to expand its existing cloud infrastructu

205、re in Singapore to meet the countrys growing customer demand for cloud technology and services.AWS also announced a new initiative called AWS AI Spring to accelerate the adoption of AI,including generative AI,in Singapore.The focus will be on developing these technologies for the public sectors,loca

206、l businesses,start-ups,communities,R&D organizations and the workforce.Currently,AWS has signed a memorandum of intent with Singapores Infocomm Media Development Authority(IMDA)for the initiative.其他國家如班、印度尼亞和來亞也在新加坡AI領域有定的投資活動。例如，Bolttech公司還從班和新加坡EDBI獲得了3000萬美元的投資。此外，新加坡也吸引了來東南亞其他國家如泰國和越南的企業前來設研發中或進

207、技術合作。Other countries such as Spain,Indonesia and Malaysia also have some investment activities in the AI field in Singapore.Bolttech,for example,also received USD30 million in investments from EDBI in both Spain and Singapore.In addition,Singapore has also attracted enterprises from other Southeast

208、Asian countries such as Thailand and Vietnam to set up R&D centers or conduct technical cooperation.7 I.Singaporehelp,support and participate in Singapores NAIS 2.0,to deepen their cooperation in AI technology,to co-create a large language model supporting 11 languages to develop Singapores homegrow

209、n AI technology,and to build a new supercomputer in Singapore.At the 10th AWS ASEAN Summit held in Singapore,Amazon Web Services(AWS)announced plans to invest an additional SGD12 billion(approximately USD8.88 billion)from 2024 to 2028 to expand its existing cloud infrastructure in Singapore to meet

210、the countrys growing customer demand for cloud technology and services.AWS also announced a new initiative called AWS AI Spring to accelerate the adoption of AI,including generative AI,in Singapore.The focus will be on developing these technologies for the public sectors,local businesses,start-ups,c

211、ommunities,R&D organizations and the workforce.Currently,AWS has signed a memorandum of intent with Singapores Infocomm Media Development Authority(IMDA)for the initiative.Other countries such as Spain,Indonesia and Malaysia also have some investment activities in the AI field in Singapore.Bolttech,

212、for example,also received USD30 million in investments from EDBI in both Spain and Singapore.In addition,Singapore has also attracted enterprises from other Southeast Asian countries such as Thailand and Vietnam to set up R&D centers or conduct technical cooperation.8 I.Singapore 2.Regulatory framew

213、ork for generative AI in Singapore1)General framework of government governance in Singapore In its National AI Strategy(NAIS),Singapore announced an ambitious goal to be at the forefront of developing and deploying scalable and influential AI solutions as a center for the development,testing,deploym

214、ent and expansion of global AI solutions.The strategy proposes five ecosystem enablers to drive the application of AI,one of which is to create a progressive and credible environment for AI,namely,to find a balance between innovation and social risks.In order to create such a progressive and trusted

215、 environment,Singapore is now adopting a more liberal and autonomous approach to AI regulation.This approach reflects Singapores two realistic considerations for AI.On the one hand,the Singapore government regards AI as an important strategic means to promote economic development and improve the liv

216、ing standards of its people.This explains why Singapore has not taken overly stringent measures in regulating AI so as not to inhibit innovation and investment.On the other hand,given its own size,Singapore recognizes that it may only be a price taker,not a price maker,in the context of evolving glo

217、bal AI governance discussions,frameworks and regulations.Therefore,instead of reformulating new AI principles,Singapore chooses to go with the world as it is,rather than what it is expected to be.Model Al Governance Framework for Generative AI(hereinafter referred to as MGF for GenAI).2024年530，新加坡政府

218、為應對成式智能技術的迅猛發展，由 AI Verify 基會和新加坡資訊通信媒體發展局（IMDA）發布MGF for GenAI。此框架建在現有政策之上，專針對如本到圖像轉換技術等成式智能的特定需求，涵蓋了九個核領域，包括問責制、數據、可信賴的開發和部署、事件報告、測試與保證、安全性、內容出處、安全與對研發（R&D）以及智能促進公共利益。其的在于實現技術優勢與潛在險的均衡，并保障AI態系統的可靠性。該框架建在2019年和2020年發布的模型 AI 治理框架和2023年發布的成式 AI：信任與治理的意義的討論件的基礎上，借鑒了與國際組織、研究社區和領先的 AI 組織的解和討論。智能治理示范框架 M

219、odel AI governance framework 示范框架（Model Framework）是新加坡在2019年世界經濟論壇年會（WEF）上推出的亞洲個智能治理示范框架，作為個愿的、強制性的框架，示范框架旨在指導各組織在規模部署智能解決案時做到負責任，不涉及技術的開發階段。因為公共部的智能使受到內部指導針和智能與數據治理具包的約束，示范框架這指南主要向私營部提供智能部署的實建議。示范框架是個“活的檔”，會根據技術和社會的變化在未來的版本中不斷更新。示范框架也不受技術、業、規模和業務模式的限制。從內容上看，示范框架遵循兩個基本原則，以增強對智能的信任和理解：第，使智能進決策的組織應該保證

220、決策過程具有可解釋性、透明性和公平性；第，智能系統應該以為本：在設計、開發和使智能時，應該優先考慮保護類的福祉和安全。The Model Framework is the first AI governance model framework in Asia launched by Singapore at the annual meeting of the World Economic Forum(WEF)in 2019 as a voluntary,non-mandatory framework.The Model Framework is designed to guide organi

221、zations to be responsible for large-scale deployment of AI solutions without involving the development phase of the technology.Because the use of AI in the public sector is governed by internal guidelines as well as AI and data governance toolkit,the Model Framework,as a guideline,is intended to pro

222、vide practical advice on AI deployment for the private sector.The Model Framework is a continuously updated document that will be updated in future releases based on technological and social changes.The Model Framework is also not limited by technology,industry,size and business model.In terms of co

223、ntent,the Model Framework follows two basic principles to enhance trust and understanding of AI:first,organizations that use AI to make decisions should ensure that the decision-making process is explainable,transparency and fair;second,AI systems should be people-oriented:in the design,development

224、and use of AI,priority should be given to protecting human well-being and safety.該框架把這些指導原則具體化為四個關鍵領域的可操作實踐，這些領域涉及組織的決策和技術的開發過程：(a)內部治理的結構和措施；(b)確定類在智能輔助決策中的參與程度；(c)業務管理；以及(d)利益相關的交流和溝通。The Framework embodies these guidelines into operational practices in four key areas involving organizational deci

225、sion-making and technology development processes:(a)structures and measures for internal governance;(b)determining the level of human participation in AI-assisted decision-making;(c)business management;and(d)stakeholder exchange and communication.下圖列出了屬于每個關鍵領域的些建議考慮因素、做法和措施的摘要。The figure below prese

226、nts a summary of some of the recommended considerations,practices and measures that fall under each key area.9 I.Singapore Model Al Governance Framework for Generative AI(hereinafter referred to as MGF for GenAI).On May 30,2024,the Singapore government responded to the rapid development of generativ

227、e AI technology,with the release of MGF for GenAI by the AI Verify Foundation and the Singapore Infocomm Media Development Authority(IMDA).The framework builds on existing policies and specifically addresses the specific needs of generative AI such as text-to-image conversion technologies,covering n

228、ine core areas including accountability,data,trusted development and deployment,incident reporting,testing and assurance,security,content provenance,safety and alignment R&D,and AI for public good.It aims to achieve a balance between technological advantages and potential risks,and to ensure the rel

229、iability of the AI ecosystem.The framework builds on the Model AI Governance Framework published in 2019 and 2020 and Discussion Paper on Generative AI:Implications for Trust and Governance published in 2023,drawing on insights and discussions with international organizations,research communities an

230、d leading AI organizations.Model AI governance frameworkThe Model Framework is the first AI governance model framework in Asia launched by Singapore at the annual meeting of the World Economic Forum(WEF)in 2019 as a voluntary,non-mandatory framework.The Model Framework is designed to guide organizat

231、ions to be responsible for large-scale deployment of AI solutions without involving the development phase of the technology.Because the use of AI in the public sector is governed by internal guidelines as well as AI and data governance toolkit,the Model Framework,as a guideline,is intended to provid

232、e practical advice on AI deployment for the private sector.The Model Framework is a continuously updated document that will be updated in future releases based on technological and social changes.The Model Framework is also not limited by technology,industry,size and business model.In terms of conte

233、nt,the Model Framework follows two basic principles to enhance trust and understanding of AI:first,organizations that use AI to make decisions should ensure that the decision-making process is explainable,transparency and fair;second,AI systems should be people-oriented:in the design,development and

234、 use of AI,priority should be given to protecting human well-being and safety.該框架把這些指導原則具體化為四個關鍵領域的可操作實踐，這些領域涉及組織的決策和技術的開發過程：(a)內部治理的結構和措施；(b)確定類在智能輔助決策中的參與程度；(c)業務管理；以及(d)利益相關的交流和溝通。The Framework embodies these guidelines into operational practices in four key areas involving organizational decisio

235、n-making and technology development processes:(a)structures and measures for internal governance;(b)determining the level of human participation in AI-assisted decision-making;(c)business management;and(d)stakeholder exchange and communication.下圖列出了屬于每個關鍵領域的些建議考慮因素、做法和措施的摘要。The figure below presents

236、 a summary of some of the recommended considerations,practices and measures that fall under each key area.10 I.Singapore of AI solutions without involving the development phase of the technology.Because the use of AI in the public sector is governed by internal guidelines as well as AI and data gove

237、rnance toolkit,the Model Framework,as a guideline,is intended to provide practical advice on AI deployment for the private sector.The Model Framework is a continuously updated document that will be updated in future releases based on technological and social changes.The Model Framework is also not l

238、imited by technology,industry,size and business model.In terms of content,the Model Framework follows two basic principles to enhance trust and understanding of AI:first,organizations that use AI to make decisions should ensure that the decision-making process is explainable,transparency and fair;se

239、cond,AI systems should be people-oriented:in the design,development and use of AI,priority should be given to protecting human well-being and safety.The Framework embodies these guidelines into operational practices in four key areas involving organizational decision-making and technology developmen

240、t processes:(a)structures and measures for internal governance;(b)determining the level of human participation in AI-assisted decision-making;(c)business management;and(d)stakeholder exchange and communication.The figure below presents a summary of some of the recommended considerations,practices an

241、d measures that fall under each key area.11 I.Singaporeunder each key area.2)Primary regulators in AI field and their responsibilities(including website,contact information,etc.)Infocomm Media Development Authority(IMDA)Website:https:/www.imda.gov.sgContact information:+65 6377 3800 The Personal Dat

242、a Protection Committee(PDPC)Website:https:/www.pdpc.gov.sgContact information:+65 6377 3131Monetary Authority of Singapore(MAS)Website:https:/www.mas.gov.sgContact information:+65 6225 5577Cyber Security Agency of Singapore(CSA)Website:https:/www.csa.gov.sgContact:1800 2550 000(for reporting crimina

243、l information)Ministry of Health(MOH)Website:https:/www.moh.gov.sgContact information:+65 6325 922012 I.Singapore(II)Core Issues of Generative AI Compliance in Singapore1.Relationship and characterization between foundation models and AI products Singapores regulatory framework does not explicitly c

244、ategorize and regulate different AI products differently,but we can still sort out the regulatory priorities to be met by different product categories from Singapores current relevant laws and regulations.In general,AI products should be mindful of complying with Singapores Personal Data Protection

245、Act(PDPA)in all cases.In addition,for large-scale foundation models and general-purpose large models,it is important to focus on complying with the macro and basic requirements for AI products in MGF for GenAI,including transparency,bias,accountability,and other issues.When it comes to model applica

246、tion,data protection should also be focused on.For large model peripheral services,if it is a cybersecurity product,attention should be paid to the requirements of the Cybersecurity Act.The large-scale foundation model is a widely trained general AI model that can perform a variety of tasks,but is n

247、ot specific to a particular industry or application scenario.Regulation of such a model focuses on transparency,data quality,bias and discrimination,and algorithmic transparency.According to the content of MGF for GenAI,transparency is the key to regulation to ensure that the training data and algor

248、ithm design of these models do not lead to unfair or discriminatory results.Data quality is also important because inaccurate or biased data can lead to models producing wrong or harmful decisions.To address these challenges,Singapore has emphasized the importance of transparency and accountability

249、in the MGF for GenAI.通模型與基礎模型類似，但這些模型經過特定領域的微調，應于更具體的任務或業。例如，在醫療領域，通模型可以于診斷疾病或預測患者的治療效果。在融領域，這些模型可以于信評估或險管理。通模型同樣適基礎模型的監管重點；另，還有通模型所在領域特定的應合規性、隱私保護和數據管理。例如，醫療應中，監管機構會特別關注模型是否符合醫療法規和患者隱私保護的要求；在融應中，重點則是模型是否遵守融監管政策并有效管理險。另外，數據安全保護為該等產品的監管重點，應確保對相關數據的處理符合PDPA。General-purpose large models are similar to

250、foundation models,but these models have been fine-tuned in specific areas for more specific tasks or industries.For example,in the healthcare sector,general-purpose large models can be used to diagnose a disease or to predict a patients therapeutic effect.In the financial sector,these models can be

251、used for credit evaluation or risk management.The regulatory focus of large-scale foundation models also applies to general-purpose large models on the one hand;on the other hand,there are specific application compliance,privacy protection,and data management in general-purpose large models domain.F

252、or example,in medical applications,regulators will pay special attention to whether the model meets the requirements of medical regulations and patient privacy protection;in financial applications,the focus is on whether the model complies with financial regulatory policies and effectively manages r

253、isk.In addition,data security protection is the regulatory focus of these products.The processing of relevant data should be ensured to be in compliance with the PDPA.模型應是指具體業或任務的應，如醫療診斷AI、融險評估AI等。這些應的監管重點在于安全性、有效性、責任歸屬以及戶同意和透明度。例如，在醫療診斷中，AI模型的安全性和有效性直接影響到患者的健康和命，因此監管機構要求這些模型經過嚴格的測試和驗證，確保其診斷結果的準確性和可

254、靠性。同時，明確的責任歸屬也是監管的重要，以確保在出現問題時可以迅速找到責任并采取相應措施。就此，同樣應該重點關注MGF for GenAI中問責制的內容。同樣，數據安全保護為該等產品的監管重點，應確保對相關數據的處理符合PDPA。Model application refers to the application of specific industries or tasks,such as medical diagnostic AI,financial risk assessment AI,etc.The regulation of these applications focuses

255、on security,effectiveness,accountability,and user consent and transparency.For example,in medical diagnosis,the security and effectiveness of AI models directly affect the health and life of patients,so regulators require these models to be rigorously tested and verified to ensure the accuracy and r

256、eliability of their diagnostic results.Meanwhile,clear attribution of responsibility is also an important aspect of regulation to ensure that the responsible party can be quickly traced and acted upon in the event of a problem.In this regard,the accountability aspect in the MGF for GenAI should also

257、 be highlighted.Similarly,data security protection is a regulatory focus for such products.The processing of relevant data should be ensured to be in compliance with the PDPA.模型周邊服務，如AI安全等，涉及提供安全保障、監控和優化AI系統等服務。絡安全局(CSA)助理席執官 Dan Yock Hau于2024年74表示，智能(AI)越來越多地嵌絡安全解決案中，以應對不斷演變的威脅。新加坡通訊及新聞部級政務部 Janil

258、Puthucheary 博強調了在絡安全具中帶頭使AI的作。對于該等絡產品，應重點確保遵守新加坡絡安全法的要求（例如，第五部分針對絡安全服務提供商的要求，未經許可，任何不得提供滲透測試和托管安全運營中（SOC）監控服務）。Large model peripheral services,such as AI security,involve providing services such as security assurance,monitoring and optimizing AI system.Dan Yock Hau,Assistant Chief Executive of the C

259、yber Security Agency of Singapore(CSA),stated on July 4,2024 that AI is increasingly embedded in cybersecurity solutions to address evolving threats.Dr.Janil Puthucheary,Senior Minister of State for Ministry of Communications and Information,highlighted the leading role of the use of AI in cybersecu

260、rity tools.For such networking products,emphasis should be placed on ensuring compliance with the requirements of the Cybersecurity Act(e.g.,Part V requirements for cybersecurity service providers,whereby no person shall provide penetration testing and hosted security operations center(SOC)monitorin

261、g services without a license).13 I.Singaporecontent of MGF for GenAI,transparency is the key to regulation to ensure that the training data and algorithm design of these models do not lead to unfair or discriminatory results.Data quality is also important because inaccurate or biased data can lead t

262、o models producing wrong or harmful decisions.To address these challenges,Singapore has emphasized the importance of transparency and accountability in the MGF for GenAI.General-purpose large models are similar to foundation models,but these models have been fine-tuned in specific areas for more spe

263、cific tasks or industries.For example,in the healthcare sector,general-purpose large models can be used to diagnose a disease or to predict a patients therapeutic effect.In the financial sector,these models can be used for credit evaluation or risk management.The regulatory focus of large-scale foun

264、dation models also applies to general-purpose large models on the one hand;on the other hand,there are specific application compliance,privacy protection,and data management in general-purpose large models domain.For example,in medical applications,regulators will pay special attention to whether th

265、e model meets the requirements of medical regulations and patient privacy protection;in financial applications,the focus is on whether the model complies with financial regulatory policies and effectively manages risk.In addition,data security protection is the regulatory focus of these products.The

266、 processing of relevant data should be ensured to be in compliance with the PDPA.Model application refers to the application of specific industries or tasks,such as medical diagnostic AI,financial risk assessment AI,etc.The regulation of these applications focuses on security,effectiveness,accountab

267、ility,and user consent and transparency.For example,in medical diagnosis,the security and effectiveness of AI models directly affect the health and life of patients,so regulators require these models to be rigorously tested and verified to ensure the accuracy and reliability of their diagnostic resu

268、lts.Meanwhile,clear attribution of responsibility is also an important aspect of regulation to ensure that the responsible party can be quickly traced and acted upon in the event of a problem.In this regard,the accountability aspect in the MGF for GenAI should also be highlighted.Similarly,data secu

269、rity protection is a regulatory focus for such products.The processing of relevant data should be ensured to be in compliance with the PDPA.模型周邊服務，如AI安全等，涉及提供安全保障、監控和優化AI系統等服務。絡安全局(CSA)助理席執官 Dan Yock Hau于2024年74表示，智能(AI)越來越多地嵌絡安全解決案中，以應對不斷演變的威脅。新加坡通訊及新聞部級政務部 Janil Puthucheary 博強調了在絡安全具中帶頭使AI的作。對于該等絡

270、產品，應重點確保遵守新加坡絡安全法的要求（例如，第五部分針對絡安全服務提供商的要求，未經許可，任何不得提供滲透測試和托管安全運營中（SOC）監控服務）。Large model peripheral services,such as AI security,involve providing services such as security assurance,monitoring and optimizing AI system.Dan Yock Hau,Assistant Chief Executive of the Cyber Security Agency of Singapore(C

271、SA),stated on July 4,2024 that AI is increasingly embedded in cybersecurity solutions to address evolving threats.Dr.Janil Puthucheary,Senior Minister of State for Ministry of Communications and Information,highlighted the leading role of the use of AI in cybersecurity tools.For such networking prod

272、ucts,emphasis should be placed on ensuring compliance with the requirements of the Cybersecurity Act(e.g.,Part V requirements for cybersecurity service providers,whereby no person shall provide penetration testing and hosted security operations center(SOC)monitoring services without a license).14 I.

273、Singapore Model application refers to the application of specific industries or tasks,such as medical diagnostic AI,financial risk assessment AI,etc.The regulation of these applications focuses on security,effectiveness,accountability,and user consent and transparency.For example,in medical diagnosi

274、s,the security and effectiveness of AI models directly affect the health and life of patients,so regulators require these models to be rigorously tested and verified to ensure the accuracy and reliability of their diagnostic results.Meanwhile,clear attribution of responsibility is also an important

275、aspect of regulation to ensure that the responsible party can be quickly traced and acted upon in the event of a problem.In this regard,the accountability aspect in the MGF for GenAI should also be highlighted.Similarly,data security protection is a regulatory focus for such products.The processing

276、of relevant data should be ensured to be in compliance with the PDPA.Large model peripheral services,such as AI security,involve providing services such as security assurance,monitoring and optimizing AI system.Dan Yock Hau,Assistant Chief Executive of the Cyber Security Agency of Singapore(CSA),sta

277、ted on July 4,2024 that AI is increasingly embedded in cybersecurity solutions to address evolving threats.Dr.Janil Puthucheary,Senior Minister of State for Ministry of Communications and Information,highlighted the leading role of the use of AI in cybersecurity tools.For such networking products,em

278、phasis should be placed on ensuring compliance with the requirements of the Cybersecurity Act(e.g.,Part V requirements for cybersecurity service providers,whereby no person shall provide penetration testing and hosted security operations center(SOC)monitoring services without a license).15 I.Singapo

279、reAct(e.g.,Part V requirements for cybersecurity service providers,whereby no person shall provide penetration testing and hosted security operations center(SOC)monitoring services without a license).16 I.Singapore 2.Compliance of large model pre-training using public data 1)Personal informationPers

280、onal data plays a vital role in building,training,and evaluating AI systems in order to enable AI models to extract user behavior patterns from these data,extract insights into user needs,and then make accurate inferences.According to PDPA,the protection of personal information is relatively compreh

281、ensive.Its regulation is not only limited to the initial collection and use of personal data by enterprises,but also extends to the entire life cycle.Therefore,if the training corpus contains large models of personal data in Singapore or involving Singapore residents,enterprises using such models mu

282、st ensure compliance with the PDPA.On 1 March 2024,Singapores Personal Data Protection Commission(PDPC)issued the Advisory Guidelines on Use of Personal Data in AI Recommendation and Decision Systems(Guidelines).The Guidelines is drafted and published based on the PDPA.While most of its requirements

283、 for the use of personal information are in line with the PDPA,it emphasizes the importance of protecting personal privacy and data security in AI systems,as well as ensuring transparency,fairness,and accountability in AI systems.The Guidelines requires that organizations using AI systems should be

284、open and transparent and include relevant practices and safeguards in their written policies to achieve fairness and reasonableness.The level of detail provided in the content of the policy should be commensurate with the risk,and organizations are generally encouraged to provide more information ab

285、out data quality during the development of the AI system and the management measures taken to safeguard the security of personal information,which may specifically include disclosure of the following information.a)數據質量：確保訓練數據集中個信息的質量(例如，數據集在市場中的代表性和最近的編制時間)，以提模型的準確性和性能采取的措施；a)Data quality:measures t

286、aken to ensure the quality of personal information in training dataset(e.g.,the representativeness of the dataset in the marketplace and the most recent compilation time)to improve the accuracy and performance of the model;b)數據匿名化及訪問限制：是否使匿化數據進模型開發，如果不是，采取了哪些流程或技術保障措施以限制只有開發員才能訪問個信息。另外，值得提的是新加坡牽頭發起的

287、東盟數據匿名化指南（ASEAN Guide on Data Anonymisation）將在2025年發布；b)data anonymization and access restrictions:is anonymized data used for model development?If not,what processes or technical safeguards are taken to limit access to personal information to developers only?In addition,The ASEAN Guide on Data Anon

288、ymisation,a Singapore-led initiative,will be released in 2025;c)偏性評估：在進偏性評估時，是否有必要使個信息來檢查特殊信息(如種族信息或宗教信息)在預訓練過程是否被使，或評估訓練數據集的偏情況；c)bias assessment:when conducting a bias assessment,is it necessary to use personal information to check whether special information(such as racial or religious informatio

289、n)is used in the pre-training process,or to assess the bias of training dataset;d)測試環境安全及訪問限制：如果模型預訓練使了個信息，企業采取了哪些程序或技術保障措施來確保測試環境的安全并限制測試員的訪問權限；d)testing environment security and access restrictions:if personal information is used for model pre-training,what procedures or technical safeguards do en

290、terprises have in place to secure the testing environment and limit tester access;e)數據最化原則：是否在智能系統開發和測試的所有階段都遵守數據最化原則等。e)data minimization principle:whether the data minimization principle is followed at all stages of the development and testing of the AI system,etc.（2）開源數據集(2)Open-source datasets除了

291、使個信息進模型訓練外，更多的企業會使市上已有的開源數據集作為模型訓練的主要數據來源，使開源數據集的可以定程度上規避知識產權、版權、個信息等帶來的險，另許多數據集有適的針對性場景，可能會更加適配所需調試的模型。當智能系統訓練使開源數據集時，企業應檢查數據集的許可證，確保數據的使為遵守其條款和條件。些開源許可證可能要求數據使者在再分發數據時保留版權聲明，共享相同許可，或在某些情況下提供源代碼。In addition to using personal information for model training,more enterprises will use existing open-sou

292、rce datasets on the market as the main data source of model training.On the one hand,using open-source datasets can avoid risks caused by intellectual property,copyright,personal information,etc.to a certain extent.On the other hand,many datasets have applicable targeted scenarios,which may be more

293、compatible with the models that need to be debugged.When using open-source datasets in the training of AI systems,enterprises should check the licenses of these datasets to ensure that the data is used in compliance with the terms and conditions.Some open-source licenses may require data users to re

294、tain copyright notices when redistributing data,to share the same licenses,or in some cases to make the source code available.前在訓練模型時較常到的開源數據集有綜合性的數據集，如UCI Machine Learning Repository，該公開數據集包含了數百個數據集，適于各種機器學習任務，如分類、回歸和聚類，還有Kaggle Datasets，這數據集提供了豐富的數據集，涵蓋了從圖像識別到本分析等多個領域。也有偏向某使場景的開源數據集，如適于圖像處理類模型的數據集

295、COCO，這數據集提供了常物體的圖像和詳細的注釋以及動駕駛領域常的開源數據集Waymo Open Dataset于動駕駛技術的研究）。還有像Reddit這樣與歌達成 6000萬美元協議，允許其利其線上討論帖訓練智能模型，形成的數據集Reddit Comments，其中包含了Reddit站上的評論數據，主要于本分析和社交絡研究。At present,some of the more commonly used open-source datasets for training large models are comprehensive datasets such as the UCI Mach

296、ine Learning Repository,which contains hundreds of datasets for a wide variety of machine learning tasks such as classification,regression,and clustering,and the Kaggle Datasets,which provides rich datasets covering a wide range of areas from image recognition to text analysis.There are also open-so

297、urce datasets that are biased towards a certain usage scenario,such as COCO,a dataset for large models in the image processing category,which provides images and detailed annotations of everyday objects,and Waymo Open Dataset,an open-source dataset commonly used in the field of automated driving for

298、 the study of automated driving technology.There are also datasets like Reddits USD60 million deal with Google that allows Google to use its online discussion posts to train AI models,resulting in the dataset Reddit Comments,which contains data on comments on Reddits site and is used primarily for t

299、ext analytics and social networking research.These datasets are often licensed in different ways.3.數據本地化和數據跨境3.Data localization and data cross-border 指南明確了PDPA如何適于AI系統。指南建議企業在AI系統開發、測試和監控中使個數據時，考慮有效性、效益性、合規性和市場性。The Guidelines identifies how the PDPA applies to AI systems.The Guidelines recommends

300、that enterprises consider validity,effectiveness,compliance,and marketability when using personal data in AI systems development,testing,and monitoring.新加坡的AI數據保護政策：Singapores AI data protection policy:智能監管模式框架：新加坡早在2019年便推出了智能監管模式框架（Model AI Governance Framework），旨在為AI開發和部署提供指導，確保數據使的透明度和安全性。Model

301、AI Governance Framework:Singapore launched the Model AI Governance Framework as early as 2019,which aims to provide guidance for AI development and deployment and ensure transparency and security of data use.AI Verify：2022年，新加坡推出了全球個智能治理測試框架和具箱AI Verify，于測試和驗證AI系統的安全性和透明度。AI Verify:in 2022,Singapore

302、 launched AI Verify,the worlds first AI governance testing framework and toolkit,to test and verify the security and transparency of AI systems.成式AI開發安全指南：新加坡計劃在2025年初推出針對成式AI開發者的安全指南，建議開發者向戶透明公開所使的數據來源、測試結果以及可能涉及的險和限制，以增強戶對AI產品的信任。Safety Guidelines for Generative AI Development:Singapore plans to l

303、aunch a safety guidelines for generative AI developers in early 2025,recommending that developers transparently and openly disclose to users the sources of data used,test results,and possible risks and limitations involved to enhance user trust in AI products.隱私增強科技：新加坡政府還持在AI領域使隱私增強科技（Privacy Enhan

304、cing Technologies），包括通過資訊通信媒體發展局監管沙盒，持與成式AI數據使相關的項。Privacy-enhancing technologies:Singapore Government also supports the use of privacy-enhancing technologies(PETs)in the AI field,including through the IMDAs regulatory sandbox to support projects related to the use of generative AI data.（1）數據本地化（1）D

305、ata localizationAI技術依賴于規模的數據處理和分析，數據本地化要求可能限制數據的由流動，影響AI模型的訓練和優化。數據本地化是指在本地存儲和處理數據的要求，旨在確保數據的主權、安全和隱私。隨著AI技術的迅速發展，數據本地化政策對AI應產了重要影響。新加坡在數據本地化有著明確的法律和政策，以保護數據隱私和確保國家安全。AI technology relies on large-scale data processing and analysis,and data localization requirements may limit the free flow of data a

306、nd affect the training and optimization of AI models.Data localization refers to the requirement to store and process data locally to ensure the sovereignty,security and privacy of the data.With the rapid development of AI technology,data localization policies have had an important impact on AI appl

307、ications.Singapore has clear data localization laws and policies to protect data privacy and ensure national security.（2）數據跨境（2）Data cross-border數據跨境傳輸要求Data cross-border transfer requirements根據新加坡個信息保護法第26條“個信息跨新加坡國境傳輸”條款規定：Under Article 26 of Singapores Personal Information Protection Act-Transmis

308、sion of Personal Information across Singapores Borders,it is stipulated that:第，對于跨境傳輸的數據，機構應當按該法律規定建個信息保護標準，確保被傳輸的數據得到與新加坡法律相等的保護，否則不得進跨新加坡國境傳輸。First,for cross-border transfer data,institutions shall establish personal information protection standards in accordance with the provisions of the Act to

309、ensure that the transmitted data is protected equal to the laws of Singapore,otherwise it shall not be transmitted across Singapore.第，新加坡個信息保護委員會可以根據機構的申請，通過書通知豁免機構前述跨境合規義務。Second,Singapores Personal Data Protection Commission(PDPC)may,upon the institutions application,waive the institutions aforeme

310、ntioned cross-border compliance obligations by written notice.第三，可豁免的情形可以由個信息保護委員會書說明；豁免不需要在政府公報中公布，并且委員會隨時可撤銷豁免。Third,exemptions may be specified in writing by the PDPC;the exemptions do not need to be published in the Government Gazette and the PDPC can revoke the exemptions at any time.第四，新加坡個信息保

311、護委員會可以隨時增加、改變或撤銷豁免的具體適情形。Fourth,the PDPC can add,change or revoke the specific applicable circumstances of the exemptions at any time.組織須遵守轉移限制的義務：組織機構在將個數據傳輸新加坡境外任何地點時，必須遵守PDPA的相關規定。組織不得將個數據轉移到新加坡以外的國家或地區，除按照PDPA規定的要求，確保被轉移的個數據將獲得與PDPA規定的保護標準相當的保護。Institutions are obliged to comply with transfer re

312、strictions:institutions must comply with the relevant provisions of the PDPA when transferring personal data to any location outside Singapore.Institutions shall not transfer personal data to a country or territory outside Singapore unless,in accordance with the requirements set out in the PDPA,it e

313、nsures that the personal data transferred will receive a standard of protection equivalent to that set out under the PDPA.要做到這點，該組織必須：For this to work,the institution must:確保其將遵守收集、使和披露個數據的義務，同時轉移的個數據仍由該組織持有或受其控制；且 Ensure that it will comply with its obligations to collect,use and disclose personal

314、data,and that the transferred personal data will still be held by or under the control of the institution;and確認且保證在新加坡境外的國家或地區的接收受到法律上可執的義務的約束，向被轉移的個數據提供少與PDPA下的保護標準相當的保護。Confirm and warrant that the recipient in a country or territory outside Singapore is bound by legally enforceable obligations to

315、 provide to the transferred personal data a standard of protection that is at least comparable to the protection under the PDPA.這些 可依法執的義務 legally enforceable obligations包括根據法律法規、合同或具有約束的公司規則（BCRs）或任何其他具有法律約束的書規定的義務。These legally enforceable obligations include those imposed under laws and regulatio

316、ns,contracts,Binding Corporate Rules(BCRs),or any other legally binding instruments.此外，持有根據個數據被轉移到的國家或地區的法律授予或認可的 特定認證specified certification 的組織將被視為受這些可依法執的義務約束。根據Personal Data Protection Regulations，特定認證 是指亞太經濟合作組織跨境隱私規則（“APEC CBPR”）體系和亞太經濟合作組織隱私識別處理（“APEC PRP”）體系下的認證。In addition,organizations tha

317、t hold a specified certification that is granted or recognized under the law of the country or territory to which personal data is transferred will be taken to be bound by such legally enforceable obligations.Under the Personal Data Protection Regulations(PDPR),a specified certification refers to ce

318、rtifications under the Asia-Pacific Economic Cooperation Cross-Border Privacy Rules(APEC CBPR)System and the APEC Privacy Recognition for Processors(PRP)System.豁免和特殊情況Exemptions and special circumstances在某些情況下，如為履合同所必需、基于個合法利益或國家利益所需，企業可以在采取合理措施后進數據跨境傳輸。此外，匿名處理后的個數據不作為個數據對待，可以由流動。In certain circumst

319、ances,such as when necessary for the performance of a contract,based on personal legitimate interests,or required in the national interest,enterprises may conduct data cross-border transfer after taking appropriate measures.In addition,anonymized personal data shall not be treated as personal data a

320、nd can flow freely.國際合作International cooperation新加坡積極參與國際協調，通過雙邊或多邊協議推動區域內形成統的數據跨境流動制度。例如，新加坡加了APEC主導的跨境隱私規則體系（CBPR），并著開發與CBPR對接的認證機制Singapore is actively engaged in international coordination to promote the establishment of a unified data cross-border flow framework within the region through bilate

321、ral or multilateral agreements.For instance,Singapore has joined the APEC-led Cross-Border Privacy Rules System(CBPR)and begun to develop an accreditation mechanism compatible with CBPR.數據跨境執法案例Data cross-border enforcement cases例如，某澳利亞物流公司新加坡主體因未通過簽署數據處理協議履數據跨境傳輸義務被處罰。此案例表明，新加坡主體在使公司統采購的境外供應商系統時，仍需

322、確保身履相關義務。For instance,a Singapore entity of an Australian logistics company was penalized for failing to fulfill its data cross-border transfer obligations by not signing a data processing agreement.This case demonstrates that even when using the overseas supplier system centralizedly procured by it

323、s parent company,a Singaporean entity must still ensure that it fulfills its own obligations.總體，新加坡的數據跨境政策較為寬松，但監管機構執法活躍，企業需要嚴格遵守相關規定以避免觸及監管紅線。Overall,Singapores data cross-border policy is relatively relaxed,but the regulatory authorities are actively enforcing the laws.Enterprises need to strictly

324、 comply with relevant regulations to avoid breaking the red line of regulation.17 I.SingaporeThe Guidelines requires that organizations using AI systems should be open and transparent and include relevant practices and safeguards in their written policies to achieve fairness and reasonableness.The l

325、evel of detail provided in the content of the policy should be commensurate with the risk,and organizations are generally encouraged to provide more information about data quality during the development of the AI system and the management measures taken to safeguard the security of personal informat

326、ion,which may specifically include disclosure of the following information.a)Data quality:measures taken to ensure the quality of personal information in training dataset(e.g.,the representativeness of the dataset in the marketplace and the most recent compilation time)to improve the accuracy and pe

327、rformance of the model;b)Data anonymization and access restrictions:is anonymized data used for model development?If not,what processes or technical safeguards are taken to limit access to personal information to developers only?In addition,The ASEAN Guide on Data Anonymisation,a Singapore-led initi

328、ative,will be released in 2025;d)c)Bias assessment:when conducting a bias assessment,is it necessary to use personal information to check whether special information(such as racial or religious information)is used in the pre-training process,or to assess the bias of training dataset;18 I.Singapore d

329、)Testing environment security and access restrictions:if personal information is used for model pre-training,what procedures or technical safeguards do enterprises have in place to secure the testing environment and limit tester access;e)Data minimization principle:whether the data minimization prin

330、ciple is followed at all stages of the development and testing of the AI system,etc.2)Open-source datasetsIn addition to using personal information for model training,more enterprises will use existing open-source datasets on the market as the main data source of model training.On the one hand,using

331、 open-source datasets can avoid risks caused by intellectual property,copyright,personal information,etc.to a certain extent.On the other hand,many datasets have applicable targeted scenarios,which may be more compatible with the models that need to be debugged.When using open-source datasets in the

332、 training of AI systems,enterprises should check the licenses of these datasets to ensure that the data is used in compliance with the terms and conditions.Some open-source licenses may require data users to retain copyright notices when redistributing data,to share the same licenses,or in some case

333、s to make the source code available.At present,some of the more commonly used open-source datasets for training large models are comprehensive datasets 19 I.Singaporedatasets for training large models are comprehensive datasets such as the UCI Machine Learning Repository,which contains hundreds of datasets for a wide variety of machine learning tasks such as classification,regression,and clusterin