CERT-In calls for testing and security checks for AI agents, focusing on how they access tools and systems, and how they prevent misuse. The guidance emphasizes verifying that agents do not use unauthorised tools, and that they respect permissions rather than escalating privileges during tasks.
The reports describe the recommendations in practical terms. The Hindu highlights the need to test agents for unauthorised tool use and privilege abuse, and it points to requirements such as checks on permissions, task delegation, and the presence of emergency-stop mechanisms. It also notes that multi-agent systems should keep authenticated, logged communications and maintain records of actions.
Times of India, in a shorter framing, also presents CERT-In’s push for checks on AI agents’ access. Together, the outlets converge on the central theme: ensuring AI agents are evaluated for safe access and operational controls before or during deployment, including auditing and accountability measures.