CERT-In calls for testing AI agents for unauthorised tool use, privilege abuse

From the publisher

Guidelines require checks on agents’ permissions, task delegation and emergency-stop mechanisms; multi-agent systems must maintain authenticated, logged communications and records of actions