Singapore studying stronger safeguards for high-risk AI uses as systems grow more autonomous: Josephine Teo

Singapore is investigating whether additional safety measures may be necessary for high-risk applications of artificial intelligence (AI), as increasingly advanced and autonomous AI systems introduce new risks beyond existing safeguards, according to Minister for Digital Development and Information Josephine Teo. In essential services and other riskier AI applications, such safeguards could comprise more comprehensive testing, evidence verifying the efficacy of safety measures, stricter deployment controls, and enhanced regulatory oversight.

Teo emphasized Singapore's commitment to fostering AI innovation while adopting a "risk-based and multi-layered approach," which involves imposing limits on AI deployment based on risk. When deploying AI agents within the public sector, the government evaluates factors like accessible systems and data, permitted actions, and potential consequences of unintended or unauthorized actions.

The Cyber Security Agency of Singapore (CSA) and Government Technology Agency of Singapore (GovTech) leverage AI to bolster the security of government systems and critical information infrastructure. This involves identifying vulnerabilities and potential threats more rapidly, as well as providing practical guidance to enhance cyber defenses and secure AI systems.

Singapore is also exploring how existing channels for reporting cyber incidents could be optimized to identify incidents involving AI, facilitate remediation efforts, and bolster safeguards. Additionally, the government is assessing the need for additional coordination or reporting arrangements. While the government does not discount catastrophic or "extinction-level" AI risks, it refrains from assuming these risks will materialize.

Instead, it closely monitors evidence, develops technical capabilities to comprehend advanced AI systems, and collaborates internationally to address severe risks as evidence emerges.

The National AI Council provides strategic guidance on the nation's broader approach to AI risks. Singapore recently endorsed an international call by more than 20 world leaders for stronger safeguards around frontier AI. The government engages frontier AI developers for information and access to their models as needed, without making such access mandatory.

According to Teo, a rigid requirement could be counterproductive by discouraging companies from engaging in information sharing. Instead, the government seeks to establish robust collaborative relationships with frontier AI labs, encouraging substantive and timely information exchange. As AI becomes increasingly capable of acting on behalf of individuals, Singapore emphasizes the necessity of introducing more robust safeguards.

Strengthening the ability to understand emerging risks and respond to problems is crucial for fostering trust in AI as a technology that serves the public good.