Singapore’s next AI industry may be the people who check the agents

Singapore's regulators have been concerned about AI agents causing liabilities, particularly when they send emails to the wrong customers with the wrong attachments after approving incorrect refunds. To address this issue, the Infocomm Media Development Authority (IMDA) announced three initiatives at the AI Assurance Exchange. These include a starter kit for testing agentic AI systems, an accreditation program for third-party AI testers, and a new taskforce to defend the infocomm and media sector against AI-enabled cyber threats.

The Starter Kit for Testing Agentic Systems for Safety and Reliability offers a four-step approach to address the unique challenges of testing agentic AI. It emphasizes identifying risks, designing realistic tests, running the tests, analyzing logs, and continuously monitoring and retesting. The kit also introduces new vocabulary like "trajectory" to examine an agent's reasoning and actions, as well as metrics such as tool use accuracy, handoff accuracy, and human-in-the-loop rates.

The AI Tester Accreditation Programme (AI TAP) aims to set a benchmark for third-party AI testing providers. Applicants undergo technical, operational, and financial evaluations, including testing competency, track record, delivery capability, and financial sustainability. The first five accredited firms include AIDX Tech, Knovel Engineering, PwC Singapore, Resaro, and Vulcan. Accreditation lasts for one year, requiring firms to maintain their status as technology evolves.

While the AI TAP is a promising move, it raises questions about regulatory requirements and potential conflicts of interest. Without regulatory mandates, the scheme could become a marketing badge rather than a safeguard. Additionally, the program's composition includes some firms that have contributed to the development and case studies, potentially creating financial auditing issues.

Nonetheless, the initiative highlights the growing importance of third-party verification for agentic AI, as 94% of Singapore businesses are using or piloting multi-step AI systems, but only 29% can produce an audit trail for AI-driven decisions.