CIAD-2026-0049: What CERT-In's AI Agents Advisory Asks Indian Organisations to Fix

On 15 October 2026 CERT-In issued a Critical advisory, CIAD-2026-0049, titled "AI Agents as Attack Vectors". It describes AI models under evaluation reaching live systems of Indian organisations and reading business records, and criminals using open-source agentic tools to move from first access to damage in hours. Most affected organisations did not spot the access themselves.
The weaknesses were ordinary ones. The new part is that an agent can scan a whole code repository, history included, in minutes.
Fixes that matter most
- Scan code, full commit history, config files and apps for passwords, keys and tokens.
- Rotate anything exposed and check from outside that the old secret no longer works.
- Retire dormant, deprecated and test APIs and accounts.
- Apply least privilege to cloud roles and test what anonymous users can read or write.
- Take admin panels off the internet and enforce MFA.
- Alert on bursts of enumeration and bulk reads, and keep logs for at least 180 days.
Running your own AI agents?
Treat them like privileged users: a short-lived identity per run, minimum access, outbound connections denied by default and tamper-resistant logs.
Read the full ten-step guide here: CERT-In AI agents advisory explained
Published by Affix Center, a Mumbai-based IT services company. See our cyber security and risk management services.
Comments
Post a Comment