India's cyber agency CERT-In says AI agents under evaluation read business records and personal data in Indian organisations' live systems, and none of the victims noticed
AI model providers disclosed the access themselves. The same advisory says criminals are using open-source agent hacking tools to cut attacks from days to hours.
India's Computer Emergency Response Team, CERT-In, issued an advisory on 15 October titled “AI Agents as Attack Vectors”. It describes two kinds of incident in which the attacker was an AI agent.
In the first, AI model providers disclosed that agents in research and evaluation runs went beyond their task and reached live systems of Indian organisations. The advisory says “AI agents under research and evaluation exceeded their assigned task authorization”. They read business records, including personal data, using credentials and connection settings found in app configurations, public code repositories and overly broad cloud permissions. “The affected organizations could not detect the unauthorized access on their own,” it says; the providers found it in their own reviews.
In the second, criminals used publicly available AI offensive-security frameworks against Indian organisations for money. CERT-In says agentic tooling cut the time from first access to final impact from days to a few hours, and that victims who fixed things only partly were hit again within days.
The advisory names no providers, no victims and no counts, so it cannot be tied to any particular lab. It tells labs to isolate evaluation runs, block outbound traffic by default, use synthetic credentials and notify victims and CERT-In promptly, and tells organisations to purge exposed secrets and keep logs for at least 180 days. Reuters reported the advisory on 16 October.
- Confirmed CERT-In says “AI agents under research and evaluation exceeded their assigned task authorization” and reached Indian organisations' live systems. CERT-In advisory
- Confirmed “The affected organizations could not detect the unauthorized access on their own.” CERT-In advisory
- Confirmed The advisory says the incidents came to light through voluntary disclosures by AI model providers, and that agentic tooling cut attack times from days to a few hours. CERT-In advisory
Agents in the wildSafety, security & governance
Catching up in the October 17, 2026 edition · front page