Singapore: New AI safety initiatives announced, including the Global AI Assurance Pilot
Singapore announced a generative AI testing pilot, a joint report with Japan and a red-teaming evaluation report to advance AI assurance and safety.
Summary
A Baker McKenzie article reports that Singapore's government revealed three AI safety initiatives on 11 February 2025, at the AI summit in France. These are a pilot on global AI assurance, a joint testing report prepared with Japan, and an evaluation report from a red teaming challenge run in Singapore. The firm says the goal is to improve AI governance, innovation and safety standards.
The assurance pilot came from the AI Verify Foundation working with the Infocomm Media Development Authority, and serves as a testbed for best practice in technical testing of generative AI applications. Providers of AI assurance and testing services will be matched with organizations that deploy generative AI. The work focuses on real applications, and lessons learned are meant to produce general guidance on what to test and how.
According to the article, results should help shape testing norms, lay groundwork for a viable AI assurance market and improve testing tools. The pilot is expected to end by May 2025 and to be presented at Asia Tech x Singapore 2025.
The joint report with Japan checks whether guardrails on large language models still work outside English. Singapore brought experts together to test across 10 languages and five harm categories. The third item examines how large language models perform across languages and cultures in the Asia Pacific and whether safeguards hold in those settings, presenting a consistent testing method. The authors conclude that the initiatives advance Singapore's governance frameworks and promote transparency and accountability as more businesses adopt AI.