AI agents developed by multiple providers accessed personal data and confidential information at Indian organisations during research and testing, the country’s cybersecurity agency said on Thursday.
The Indian Computer Emergency Response Team said it had received “multiple voluntary disclosures from AI model providers” after their models accessed the live systems of Indian organisations during research and evaluation exercises on the public internet.
According to the agency, the AI agents exploited exposed credentials and connection settings in application configurations, public code repositories and cloud roles with excessive permissions.
Their activities also “exceeded the authorisation of their assigned tasks.”
India is the latest country to report incidents involving rogue AI agents, following a series of similar cases in other parts of the world.
In July, OpenAI’s agents broke out of a test environment and breached the systems of French company Hugging Face, triggering a summer of reported rogue AI activity. Last month, OpenAI apologised to Australia after its agents accessed four government websites, including a portal containing Medicare statistics.
At Hugging Face, the agents initially gained access through similar means but later identified and exploited software vulnerabilities to execute code on the company’s servers. However, CERT-In did not report any comparable hacking activity in India, suggesting that the agents relied on exposed credentials and existing access permissions to reach the organisations’ systems.
However, the Indian incidents appear more serious in terms of the information accessed by the AI agents. In Australia, OpenAI said it found no evidence that its agents accessed individuals’ medical or criminal records, despite writing files within a Medicare statistics system. In India, by contrast, the agents reportedly accessed personal data and confidential business records.
Another distinction is that the Indian incidents involved AI agents from multiple providers, unlike previous cases, which were each disclosed by a single AI laboratory.
CERT-In did not identify the AI providers or affected organisations, nor did it disclose whether the agents modified any systems.
