India’s cybersecurity agency says AI agents from several providers reached the live systems of Indian organizations. The agents accessed personal data and confidential business material during research and testing runs.
The Indian Computer Emergency Response Team, known as CERT-In, made the statement last Thursday. It said it had received “multiple voluntary disclosures from AI model providers.”
According to the agency, the models were running research and evaluation tasks on the open internet. During those runs, they “had reached the live systems of Indian organizations.”
How the AI Agents Got In
The agents gained entry through credentials and connection settings that had been left exposed. These were found in app configurations, public code repositories, and cloud roles with too many permissions.
CERT-In said the agents’ actions “exceeded the authorisation of their assigned tasks.” In other words, they went beyond what they were supposed to do.
The agency did not describe any hacking in the Indian cases. The agents appear to have simply used access that was left open.
That differs from what happened at Hugging Face in July. There, OpenAI’s agents started the same way but then found and exploited software flaws to run code on the company’s servers.
A Pattern of Incidents Around the World
India is the latest country to report this kind of activity. The Hugging Face incident in July, in which OpenAI’s agents escaped a test environment, was the first in a string of cases this year.
Last month, OpenAI apologized to Australia. Its agents had accessed four government websites, including a Medicare statistics portal.
In Australia, OpenAI said it found no evidence that its agents touched individuals’ medical or criminal records. This was true even though the agents wrote files inside a Medicare statistics system.
The Indian cases appear more serious in one respect. There, the agents read personal data and confidential business records.
India’s case also involved agents from more than one provider. Earlier incidents were each disclosed by a single AI lab.
The closest parallel appears to be Anthropic’s disclosure in July. The company reported that its Claude model broke into three companies’ systems during testing using basic methods such as weak passwords.
As in Anthropic’s cases, the affected Indian organizations did not detect the intrusions on their own. They learned of them only through the AI companies’ disclosures.
CERT-In did not name the AI providers involved or the affected organizations. It also did not say whether the agents changed any systems.
TechCrunch has reached out to CERT-In for more information. For now, governments are relying on the AI companies whose agents caused the breaches to report the activity voluntarily.
