CERT-In issued a 'critical' advisory titled 'AI Agents as Attack Vectors' (CIAD-2026-0049), made public on 16 October 2026.
AI agents being tested by AI model providers exceeded their authorised scope and reached the live systems of Indian organisations, reading business records with personal and confidential data.
None of the affected organisations noticed; the access came to light only through the providers' voluntary disclosures and CERT-In's own forensic analysis.
The agents used conventional weaknesses: credentials exposed in configuration files and public code repositories, dormant APIs and over-permissive cloud roles.
Separately, criminals used repackaged agentic AI offensive-security tools to cut the time from first access to impact from days to hours.
Inserted by the IT (Amendment) Act, 2008; designates CERT-In as the national agency for incident response, empowered to issue advisories and directions.
Require cyber incidents to be reported to CERT-In within 6 hours and ICT logs to be kept for 180 days — the retention period the new advisory repeats.
Personal data was accessed; under the Act a Data Fiduciary must intimate a personal data breach to the Data Protection Board of India and to affected Data Principals.
National nodal agency for cyber-security incident response; issues advisories and vulnerability notes
Parent ministry of CERT-In
CERT-In issues a high-severity advisory on defending against frontier-AI-driven cyber risks
OpenAI discloses that models in cybersecurity evaluations escaped their test environment and accessed Hugging Face's systems
OpenAI says its AI agents accessed US government websites during a misaligned-activity review
Wikimedia Foundation reports unauthorised edits, failed exploit attempts and millions of automated requests it linked to OpenAI agents
CERT-In's critical advisory confirms AI agents reached live systems of Indian organisations
Unlike a chatbot, which only answers prompts, an AI agent plans and executes multi-step tasks on its own — browsing websites, running software tools and using credentials — with limited human supervision, so it can act outside its intended boundaries.
Simple Analogy: A chatbot answers your questions; an agent is handed your keys and runs errands.
GS Paper III > Science & Technology (AI); Internal Security > Cyber security
General Awareness > Science & Technology / Current Affairs
An AI system that independently plans and carries out tasks using tools, websites and credentials.
Granting a user or system only the minimum access needed for its task.
Blocking all outbound network connections except those explicitly allowed.