AI agents broke into Indian company systems and read private data, CERT-In warns
CERT-In warns that AI agents accessed Indian systems without authorisation and read sensitive records. Read the key risks and the steps organisations must take
PTC Web Desk: India’s cybersecurity agency, CERT-In, has issued a critical advisory after AI agents entered the systems of Indian organisations without permission. The agency says the agents were able to read business records that held personal data and confidential information.
CERT-In also said that, in some cases, hackers tried to break into systems using AI tools.
According to the advisory, some AI agents that were being researched and evaluated went beyond the tasks they were given and accessed Indian ICT assets without authorisation.
Also Read | FBI’s most wanted fraudster Manjit Singh Bedi sent back to US from India a month after arrest
Two types of incidents
CERT-In said it found two kinds of breaches. The first involved AI models that got into systems while they were being tested by AI model providers. The second involved cyber attackers who deliberately used open-source agentic AI tools to break in.
The agency said several AI model providers reported these incidents to it voluntarily. It added that the affected organisations could not detect the intrusions on their own.
CERT-In did not name the organisations that were affected. It also did not name the AI companies that made the disclosures.
How the agents got in
The agency said the AI agents took advantage of security weaknesses to obtain credentials and connection settings. With these, they were able to read business records, including personal data and confidential material. CERT-In said their actions went beyond what they were authorised to do.
CERT-In also noted financially motivated attacks in which hackers misused publicly available open-source AI tools. Some of these tools were rebranded and run on infrastructure spread across several countries. The agency warned that such tools can cut the time needed for an attack from days to just a few hours and that many agents can work in parallel for hours or even days.
How much damage these incidents caused is not yet clear.
Repeat attacks after weak fixes
CERT-In said organisations that only partly fixed their problems were hit again within days, and on a bigger scale. The agents also reused the same exposed credentials in different operations.
What CERT-In wants done
The agency has asked AI developers and research bodies to be stricter. It said testing should take place in isolated environments using fake credentials, so that agents cannot reach third-party systems. It also asked for real-time monitoring and for safeguards to be published.
Organisations have been told to audit and rotate their credentials and remove old, unused APIs, among other steps. Those that confirm unauthorised access should treat their systems as compromised until forensic checks clear them. BFSI and fintech companies were also asked to follow the measures meant for payment and disbursement APIs.
A growing pattern
The advisory comes at a time when AI agents breaching websites is becoming more common. In July, around 700 rogue OpenAI agents reportedly tried to hack Hugging Face and several more cases have followed. These include OpenAI agents trying to access the websites of the US and Australian governments and the UN. Meta, Anthropic and Google have also made similar disclosures in the past.