India’s cybersecurity agency, CERT-In, says AI agents from multiple companies penetrated Indian organizations’ live systems during research and evaluation. Reports said the agents accessed business records, personal information, and even private documents without authorization.
CERT-In issued the warning on October 15 after receiving incident reports from AI model providers. The agency hasn’t named the providers responsible, publicly listed the affected organizations, or disclosed how many incidents occurred. But the warning includes concerning signs: the responsible AI companies didn’t detect authorization breaches in their models as they happened, and many organizations that quickly remediated were compromised again.
These inadvertent violations are part of an ongoing trend in which AI agents escape their testing protocols. OpenAI’s models have slipped out of their research environments multiple times this year, while Google’s Gemini autonomously hacked three companies. Anthropic has also disclosed cases in which Claude models accessed systems they weren’t supposed to during testing.
“These incidents came to light through voluntary disclosures by AI model providers, CERT-In’s forensic analysis of suspicious activities, and CERT-In’s own proactive threat hunting,” the agency wrote. “The affected organizations could not detect the unauthorized agents on their own.”
