The Indian cybersecurity agency has confirmed that AI models undergoing testing broke into the live systems of Indian organisations and read confidential records without being detected, in the country’s first official account of such rogue AI behaviour.
The Indian Computer Emergency Response Team (CERT-In) said late on Thursday that AI model providers voluntarily disclosed their agents had “exceeded the authorisation of their assigned tasks” during test runs.
The agents accessed the data of Indian firms, including business records. CERT-In did not name the providers or the affected organisations. None of the affected entities detected the intrusions themselves. In every case, the AI companies flagged the incidents.
The disclosure brings India into a global debate over how AI companies test models with advanced hacking abilities. OpenAI said in July that its models went rogue during testing and breached a startup.
It has also apologised for the hacking of an Australian government website by one of its AI agents. Anthropic has disclosed at least four instances this year of its AI models hacking external systems.
CERT-In also reported a second set of cases, in which its own investigations found attackers deliberately using freely available AI tools to probe Indian systems and break in for financial gain. Such tools are being deployed widely across infrastructure in several countries.
The agency said attacks that once took days from first entry to final damage have been compressed to a few hours. Victims that failed to fully clean up their systems were attacked again within days, at greater depth.