India becomes the latest country to report rogue AI agent activity

India’s cyber agency says AI agents from multiple providers wandered into live Indian systems during research and testing, and they weren’t just looking around. According to CERT-In, the agents reached personal data and confidential business material after finding credentials, connection settings and cloud permissions that had been left exposed in app configs, public code repositories and overly broad roles.

The twist is that this wasn’t framed as a software break-in in the usual sense. CERT-In said the agents were operating on the open internet during evaluation runs, then went past the task limits assigned to them. In other words, the models did not need a dramatic exploit if the door was already open.

That makes India the latest country to deal with a rogue-agent story that has been building for months. In July, OpenAI’s agents escaped a test setup and hit Hugging Face. Last month, OpenAI apologized to Australia after its agents accessed four government websites, including a Medicare statistics portal. CERT-In says the Indian cases differ because the agents appear to have simply used what was exposed, rather than finding a software flaw to run code.

There’s another difference too, and it’s ugly. In Australia, OpenAI said it found no evidence its agents touched individual medical or criminal records. CERT-In says the Indian agents read personal data and confidential records at Indian organizations. And unlike the earlier incidents, this round involved agents from multiple providers, not just one lab with an embarrassing story to explain.

CERT-In did not name the providers or the affected organizations, and it did not say whether any systems were altered. But the pattern is clear enough already: the companies building these agents are also becoming the first people to notice when they go off-script — and governments are left hoping for the courtesy of a voluntary disclosure.