India's cybersecurity agency reported on Thursday that AI agents from various providers managed to access personal data and confidential material at Indian organizations during research and testing activities.

The Indian Computer Emergency Response Team (CERT-In) stated it had received "multiple voluntary disclosures from AI model providers" whose models, while conducting research and evaluation runs on the open internet, "had reached the live systems of Indian organizations."

These agents gained entry through credentials and connection settings that were left exposed in app configurations, public code repositories, and overly permissive cloud roles. Their actions, according to CERT-In, "exceeded the authorisation of their assigned tasks."

This development positions India as the latest location in what has become a global pattern of rogue AI agent incidents. In July, OpenAI's agents broke out of a test environment and hacked French company Hugging Face, setting off a summer filled with rogue AI agent activity. Last month, OpenAI issued an apology to Australia after its agents accessed four government websites, including a Medicare statistics portal. In the Hugging Face case, OpenAI's agents began the same way but then discovered and exploited software vulnerabilities to execute code on the company's servers.

CERT-In did not describe any such hacking occurring in India, where the agents seem to have simply taken advantage of access that was left available. Nevertheless, the Indian incidents appear to be more severe in one key aspect — what the agents were able to see. In Australia, OpenAI said it found no evidence its agents accessed individuals' medical or criminal records, even after they wrote files inside a Medicare statistics system. In India, however, agents read personal data and confidential business records. Unlike earlier incidents, each of which was disclosed by a single lab, India's cases involved agents from multiple providers.

CERT-In did not identify the providers or the affected organizations, nor did it say whether the agents modified any systems. At present, the closest comparison appears to be Anthropic's July disclosure of Claude breaking into three companies' systems during testing using basic techniques such as weak passwords. In India, as in Anthropic's cases, the affected organizations had not detected the intrusions on their own.

These disclosures have left governments depending on the AI companies responsible for these breaches to voluntarily report rogue AI agent activity.