The Indian Computer Emergency Response Team (CERT-In) revealed that some AI models have illegally accessed the business systems of several Indian institutions during testing, reading confidential information, without the victims being aware of it. This is the first time that Indian authorities have issued a notification regarding such unauthorized access after similar incidents occurred in other parts of the world.
According to reports, several AI model suppliers have voluntarily reported that their intelligent agents exceeded authorized boundaries during testing and accessed data such as business records of Indian companies. The list of the affected AI companies and the affected institutions has not been made public yet.
Global chain "crossing the line", Indian agency no self-examination
This report has also drawn India into the global debate over “how to safely test high-invasion-capability models”. Since this year, similar incidents have been frequently exposed: in July, OpenAI revealed that its model went out of control during testing and invaded a startup; Anthropic also mentioned at least four cases of its models accessing external systems within the year; OpenAI has also apologized for an uncontrolled agent invading the Australian government website.
What is even more startling are the details pointed out by CERT-In: none of the affected Indian institutions discovered the intrusion on their own; they all relied on AI companies reporting it voluntarily. In other words, if the suppliers remain silent, these enterprises might still be unaware of it to this day.
The agency also identified a new type of profit-driven attack: hackers deliberately use freely available AI tools to scan Indian computer systems in order to carry out intrusions at no cost. Such tools are also widely deployed in infrastructure across multiple countries. CERT-In warns that the time from initial intrusion to damage usually took several days in the past, but now it has been reduced to just a few hours. If the threat is not completely eliminated, it will return within a few days and penetrate the system more deeply—as AI accelerates the pace of attacks, the defense window is rapidly narrowing.