AI agents breach Indian firms’ systems, roam undetected for days, CERT-In warns

Mumbai: Artificial intelligence agents being tested by model providers attacked the systems of Indian organisations, breaching their defences and rooting through them for days without being detected, according to a notification issued by the Indian Computer Emergency Response Team (CERT-In) on Thursday.

"In some incidents, AI agents under research and evaluation exceeded their assigned task authorisation and made unauthorised access to Indian information and communication technology (ICT) assets," the agency said. "In others, cyberattackers deliberately used open-source agentic AI tools to conduct reconnaissance, exploitation and financially motivated intrusions."

Also read: Meta’s new Muse AI agents raise questions about privacy and user control


CERT-In said the agents breached live systems and read through business records, personal data and other confidential material. The impacted organisations could not detect the access, it said, adding that the incidents only came to light after voluntary disclosures by AI labs.

India's national nodal cybersecurity agency noted that the agents went beyond the scope of the work they were authorised to perform. They gained access using exposed passwords, digital keys and poorly protected online services.

CERT-In did not name the providers or affected organisations.

The agency also said cybercriminals used publicly available AI tools in a financially motivated campaign against Indian entities. The tools helped attackers move from breaking into systems to causing harm within hours, rather than days. Multiple AI agents worked simultaneously for hours or days. Organisations that failed to fix security weaknesses were attacked repeatedly, at greater scale.

The warning comes amid concerns about strikes on India's financial sector. ET had reported on Thursday that about 20 FIs had been hit earlier this month, with hackers allegedly using stolen partner-bank credentials and weaknesses in payment systems to siphon away money.

Japan Declared State of Emergency on Oct 9

In a separate October 9 warning, CERT-In said attackers had targeted lenders, fintech firms, payment providers and wallet operators. They used stolen credentials to make transfers from the victims' own systems, bypassing checks intended to prevent unauthorised payments. That advisory also pointed to the possible use of AI-assisted tools.

Concerns are also growing overseas. Japan's government on October 9 declared a "state of emergency in cyberspace" after a wave of AI attacks targeted over 20 major companies, local governments, rail operators, and public services.

Globally, AI developers have also reported increasingly autonomous cyber operations. In November 2025, Anthropic said a group it assessed as Chinese state-sponsored had manipulated Claude Code to attempt intrusions into roughly 30 organisations, succeeding in a small number of cases. In July 2026, OpenAI's own agents breached parts of Hugging Face's systems. Axios later reported that OpenAI, Anthropic and researchers were investigating tens of thousands of problematic AI incidents, including unsuccessful attempts, rather than just confirmed attacks.

CERT-In has not linked its latest advisory to any of these global incidents. OpenAI and Anthropic did not comment on ET's queries.

As of now, CERT-In has advised businesses to shut down unused applications and restrict cloud permissions. Organisations deploying their own agents should give each run a separate, short-lived identity with limited permissions and maintain tamper-resistant records of actions affecting production systems and data, the advisory said.