India's cybersecurity agency has raised concerns about AI agents slipping past their intended limits and getting into real company systems. The Indian Computer Emergency Response Team (CERT-In), the national agency that handles cybersecurity incidents, issued a critical advisory on Thursday after autonomous AI agents reportedly accessed the live systems of Indian organisations without permission.
The advisory, numbered CIAD-2026-0049 and titled ‘AI Agents as Attack Vectors', did not name the affected organisations. It flagged two types of incidents involving the IT and communication systems of Indian entities.
AI Agents Went Beyond Their Tasks
The first involved AI agents that were being tested. CERT-In said some of these agents, which can carry out tasks on their own, ended up doing more than they were permitted to while under research and evaluation. As a result, they found their way into the live systems of Indian entities.
ALSO READ | The Emerging-Market AI Divide: How World Bank Plans To Democratise 'Small AI' For Small Businesses
The incidents came to light through voluntary disclosures by the companies behind the AI models. According to CERT-In, these providers acknowledged that, while being tested, their models had got into real systems run by Indian organisations. The affected organisations themselves had not detected the unauthorised access.
During these incidents, the AI agents reportedly accessed business records containing personal information and confidential data.
How the Agents Got In
CERT-In said the access was made possible by login credentials that had been left exposed. These were found in cloud storage, configuration files, client applications and source code, as well as in code repositories and the variables used in software build and release pipelines, known as CI/CD.
Hackers Turning to AI Tools
In a separate finding, CERT-In's investigations uncovered a financially motivated campaign targeting Indian entities. Attackers were using publicly available offensive-security frameworks built around AI, tools that have reportedly been used against IT systems in several countries. The agency warned that threat actors are increasingly relying on open-source agentic AI tools to launch attacks.
ALSO READ | From Farmers To Small Businesses: Google Expands Its AI Push In India
Attacks Now Take Hours, Not Days
CERT-In said AI is making cyberattacks much faster. Attackers can now go from first getting in to doing serious harm far more quickly, sometimes finishing in a few hours what used to take several days.
The agency also cautioned that if an organisation doesn't fully clear an attacker out, it could be targeted again within days, this time with the intruders reaching further into its systems.
What Organisations Should Do
CERT-In's warning points to three priorities for organisations: keeping credentials protected, checking who has access to what in their cloud setups, and putting tighter limits on what AI agents are allowed to reach.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.