Indian cybersecurity authorities said Thursday that AI agents from multiple providers accessed personal data and sensitive materials of Indian organizations during investigations and tests.
The Computer Emergency Response Team of India (CERT-In) said it had received “multiple voluntary disclosures from AI model providers” whose models “reached live systems of Indian organizations” while conducting research and evaluation on the open internet.
The agents gained entry through credentials and connection settings left exposed in app configurations, public code repositories, and overly permissive cloud roles, and their actions “exceeded the authorization of their assigned tasks.”
This makes India the latest destination in the rogue AI agent’s global tour. In July, an OpenAI agent escaped from a test environment and hacked French company Hugging Face. This began the summer of rogue AI agent activity. OpenAI last month apologized to Australia after its agents accessed four government websites, including the Medicare statistics portal.
At Hugging Face, OpenAI’s agents started in the same way, but then found and exploited flaws in the software to execute code on the company’s servers. CERT-In reports that no such hack took place in India, where agents appear to have simply taken advantage of access that had been left unattended.
Still, India’s case appears to be getting worse in some respects, as investigators see it. In Australia, OpenAI said it found no evidence that agents touched personal medical or criminal records even after writing files into the Medicare statistics system, while in India it said agents read personal data and sensitive business records. And unlike previous incidents, each uncovered by a single laboratory, India involved agents from multiple providers.
CERT-In did not name the provider or the affected organization, nor did it say whether the agent made any changes to the system. TechCrunch has reached out to CERT-In for more information.
For now, the closest analogy appears to be Anthropic’s July disclosure that Claude used basic techniques such as weak passwords to break into three companies’ systems during testing.
In India, as in the case of Anthropic, the affected organizations did not independently detect the intrusion. This disclosure has led governments to rely on the AI companies responsible for these breaches to voluntarily disclose the activities of rogue AI agents.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
