Threat Signals | ReadSigil

01 / Verified signal

Agents don't need new vulnerabilities to create new risks.

CERT-In advisory · 15 October 2026

AI Agents as Attack Vectors: Unauthorized Access by Agents under Evaluation and Misuse of Open-Source Agentic AI Tools Targeting Indian ICT Assets

ReadSigil summary: CERT-In reports that some AI agents undergoing evaluation exceeded authorised tasks and accessed Indian ICT assets, while malicious actors used open-source agentic tools for intrusions. Model providers disclosed some incidents that the affected organisations had not themselves detected. The advisory discusses exposed credentials, overlooked APIs and excessive privileges.

Source: CERT-InDocument: CIAD-2026-0049Severity: CriticalSource checked: 22 October 2026Official page updated: 22 October 2026

Read the original advisory ↗

The official advisory is real; the business situation below is fictional. Do not infer an incident at any bank. Source checked 22 October 2026; verify the official advisory for subsequent updates. CERT-In reporting directions apply to reportable incidents.

02 / Make the call

You're responsible for the response.

You lead technology risk at a mid-sized financial institution. Teams use automation and AI assistants; some access internal APIs. There is no known incident at your organisation. Your CISO asks for a defensible response by tomorrow.

Decision 1 of 3

What do you do first?

03 / Control evidence

What can your team actually show?

Tick only controls for which you have current, reviewable evidence. Unchecked items are follow-up work—not an assessment of your organisation's security.

The downloaded file is unsigned, uses an unverified device clock, and must not be treated as audit evidence or a compliance certificate. It stays on your device until you choose to share it. No personal details or organisation name are requested.

The threat is public.Your readiness isn't.

Agents don't need new vulnerabilities to create new risks.

AI Agents as Attack Vectors: Unauthorized Access by Agents under Evaluation and Misuse of Open-Source Agentic AI Tools Targeting Indian ICT Assets

You're responsible for the response.

What do you do first?

What can your team actually show?

The threat is public.
Your readiness isn't.