IMDA launches three initiatives for AI safety and resilience
One key measure is a starter kit to test AI agents.
Singapore is rolling out three initiatives to strengthen artificial intelligence (AI) safety, including a testing guide for AI agents, accreditation for testing firms, and a task force on AI-driven cyber threats.
The Infocomm Media Development Authority announced the measures at the AI Assurance Exchange: The Agentic Edition on 14 October, amidst risks in using agentic AI.
The Starter Kit for Testing Agentic Systems for Safety and Reliability provides a practical reference for organisations to test AI agents, which can make changes to systems and access sensitive data on behalf of users.
The kit uses a four-pronged approach, comprising identifying risks, creating realistic tests, running tests and analysing logs, and monitoring and re-testing.
It draws on consultations with government agencies and over 20 industry players, including Google, AWS, and Resaro.
The kit will be updated as testing practices evolve.
Meanwhile, AI Verify Foundation launched the AI Tester Accreditation Programme for third-party AI testing service providers.
The programme requires firms to undergo technical, operational, and financial evaluations for qualification. Accreditation lasts one year, allowing regular reviews as AI develops.
The first five accredited firms are AIDX Tech, Knovel Engineering, PwC Singapore, Resaro, and Vulcan.
The Frontier AI Resilience and Security Taskforce will publish a report with practical guidelines by the first half of 2027 regarding response, resilience, and intelligence on AI attacks.
The task force brings together more than 20 representatives and experts from the telecommunications, media, cybersecurity, and AI sectors.