Introduction
The rapid advancement of Artificial Intelligence (AI), particularly General Purpose AI (GPAI), necessitates comprehensive safety research to ensure these systems are powerful [1], safe [1] [4], reliable [1], and aligned with human values [1]. The Singapore Consensus on Global AI Safety Research Priorities [1] [2] [4], developed by over 100 AI experts from 11 countries [1], serves as a central foundation for organizations developing or deploying AI. It establishes a shared research agenda that connects AI science with policymaking through structured dialogue, identifying critical areas of AI safety requiring international focus [1].
Description
As AI technologies [1], particularly General Purpose AI (GPAI), continue to advance rapidly, there is an urgent need for comprehensive safety research to ensure these systems are powerful [1], safe [1] [4], reliable [1], and aligned with human values [1]. The Singapore Consensus on Global AI Safety Research Priorities [1] [2] [4], developed by over 100 AI experts from 11 countries [1], serves as a central foundation for organizations developing or deploying AI. It establishes a shared research agenda that connects AI science with policymaking through structured dialogue, identifying critical areas of AI safety requiring international focus [1], including risk assessment, the development of trustworthy systems, and post-deployment control mechanisms.
Risk assessment remains a core research priority [1], focusing on methods to measure and predict potential harm from AI systems. This includes standardized audit procedures and precise benchmarks for dangerous capabilities, as well as enhancing metrology for accurate assessments. Trustworthy AI systems are essential for public confidence [1], necessitating a clear understanding of desired behaviors and their technical implementation to ensure safety. Continuous monitoring and emergency protocols are vital for addressing unforeseen behaviors after deployment.
The AI Verify framework [3], developed by the Infocomm Media Development Authority (IMDA) in collaboration with the US National Institute of Standards and Technology (NIST), consists of 11 governance principles designed to validate AI system performance through standardized tests [3], including transparency [3], explainability [3], safety [1] [2] [3] [4], security [3], fairness [3], and accountability [1] [3]. Initially released in May 2022 [3], AI Verify garnered interest from over 50 companies and was open-sourced in 2023 to address risks associated with Generative AI [3]. The AI Verify Foundation [3], comprising major members such as AWS, Google [3], and Microsoft [3], guides the development of testing tools for responsible AI use [3]. In February 2025 [3], the Foundation and IMDA initiated the Global AI Assurance Pilot to establish norms and best practices for testing Generative AI applications [3].
The International AI Safety Report (IAISR) provides a comprehensive overview of the risks associated with GPAI, ranging from targeted misuse to unintended malfunctions. Specific risks arise from malicious use, such as the generation of deepfakes or support for cyberattacks, while systemic risks affecting societal areas like privacy and copyright present additional challenges. The complexity of risk assessment is heightened by the multitude of unpredictable use cases and the lack of standardized evaluation criteria.
Research also emphasizes monitoring and intervention mechanisms for AI systems [1], extending oversight to the broader AI ecosystem and enhancing societal resilience to adapt to AI-related changes [1]. A globally aligned research roadmap is crucial for developing local systems in line with international best practices [1], fostering collaboration among nations and regional groups like ASEAN [1]. Capacity building is vital for engaging research institutions and universities in advanced safety research [1]. As Southeast Asia rapidly adopts AI across various sectors [1], ensuring trust in these systems is paramount [1]. A holistic approach to building trustworthy AI is necessary [1], addressing governance areas to facilitate seamless AI adoption in the region [1].
Singapore’s Model AI Governance Framework [3], first published in January 2019 and updated in January 2020 [3], provides ethical and governance guidance for private sector AI deployment [3]. The AI Verify Foundation and IMDA have since expanded this framework to include considerations for Generative AI [3], supported by the ISAGO guide for assessing alignment with governance practices and a Compendium of Use Cases showcasing practical implementations [3]. Additionally, IMDA [3], in collaboration with the Lee Kuan Yew Centre for Innovative Cities [3], has launched a guide to help organizations redesign job roles to leverage AI’s potential [3], addressing its impact on employees [3]. The Trusted Data Sharing Framework has also been introduced to facilitate trusted data-sharing partnerships [3], with IMDA engaging stakeholders to shape Singapore’s AI ecosystem [3].
Singapore has taken significant steps to foster a trusted AI ecosystem in ASEAN [1], chairing the AI Governance Working Group and coordinating AI projects [1]. The ASEAN Guide on AI Governance and Ethics provides practical frameworks for responsible AI design and deployment [1], promoting alignment and interoperability across jurisdictions [1]. Furthermore, a Joint Statement on AI Safety was signed with France’s AI Safety Institute [4], further solidifying global partnerships in this critical area [4].
To represent Southeast Asian perspectives in global AI developments [1], Singapore organized a regional AI Safety Red Teaming Challenge [1], involving experts from ASEAN member states and dialogue partners [1]. These initiatives aim to integrate diverse viewpoints into global discussions on AI safety [1].
The Singapore Consensus is intended to be a dynamic document [1], evolving with technological advancements and stakeholder feedback [1]. It serves as a foundation for ongoing regional and global dialogue on advancing AI safety through cooperation and shared objectives [1], with a commitment to ensuring AI developments benefit the global public good [1].
Conclusion
The initiatives outlined in the Singapore Consensus and related frameworks underscore the importance of international collaboration in AI safety research. By establishing shared research agendas and governance frameworks, these efforts aim to mitigate risks associated with AI technologies while fostering public trust. The proactive measures taken by Singapore and its partners highlight the critical role of structured dialogue and cooperation in ensuring that AI advancements contribute positively to society, ultimately benefiting the global public good.
References
[1] https://oecd.ai/en/wonk/strengthening-global-ai-safety-a-perspective-on-the-singapore-consensus
[2] https://www.datenschutzticker.de/2025/05/ki-sicherheit-singapore-consensus-fordert-mehr-forschung-zur-kontrolle/
[3] https://www.imda.gov.sg/about-imda/emerging-technologies-and-research/artificial-intelligence
[4] https://techcoffeehouse.com/2025/05/28/singapores-ai-ambitions-take-centre-stage-at-atxsummit/
