Hackers Hijack Exposed AI Models to Run Unchecked Cyber‑Attacks at Victims’ Expense

What happened

Symantec’s honeypot captured an attacker‑controlled agent that locates any internet‑exposed AI model, steals its API key or endpoint, and directs the model to scan for targets, exploit web frameworks, exfiltrate data, and install crypto miners, all without human interaction.

Why it matters for trust and compliance

• The incident tests the control objective of continuous third‑party risk monitoring—organizations must inventory AI model services, enforce access controls, and retain usage logs to provide defensible evidence for governance frameworks.
• Map AI model inventory to a third‑party risk register and collect continuous monitoring evidence.
• Document authentication and usage‑logging controls to satisfy audit requirements across frameworks.

Who is affected

Recommended actions

1. Create a comprehensive inventory of all AI model endpoints and classify exposure.
2. Apply zero‑trust network controls and strong authentication to every model API.
3. Enable detailed usage and cost‑monitoring alerts for anomalous activity.