Singapore’s Infocomm Media Development Authority has announced a starter kit for testing AI agents before and during deployment. Its four steps cover risk identification, test design, results and activity logs, and continued monitoring. The kit examines the actions an agent takes as well as its final answer, giving organisations guidance for systems that use tools and access sensitive information.
Singapore’s Infocomm Media Development Authority, the regulator known as IMDA, announced a starter kit for testing AI agents on 14 October. It is intended to help organisations assess safety and reliability when deploying systems that can execute tasks on a user’s behalf. The new guidance considers the steps taken during a task alongside the final output.[1], [2]
The process begins with identifying risks and designing tests. Organisations then analyse results and activity logs, followed by continuous monitoring and retesting. These checks address agents that use tools, access sensitive information and perform operations for users. The guide is meant to support preparation before deployment as well as continued assessment once the system is in use.[1]
IMDA presented the guide at the AI Assurance Exchange, an event in Singapore on assurance for AI systems. It draws on work by the Global AI Assurance Sandbox, a programme for testing AI, and the Singapore AI Safety Institute, which evaluates AI safety. More than twenty industry participants were consulted. IMDA chief executive Ng Cher Pong announced the initiative as part of the authority’s work on agent safety and cyber resilience.[1]
