Lab: Exploiting AI agents to perform destructive actions
Summary: PortSwigger Research published a lab in which an AI-powered scanner, given login credentials for the user carlos, investigates user-generated content on a site. The lab's goal is to delete carlos by using indirect prompt injection to manipulate the LLM's behavior through third-party content.
What you should know: The scanner is handed real credentials for carlos and can be steered by content it reads, so an agent with authenticated access can be turned against the account it was trusted to inspect.
Read the original at PortSwigger Research ↗
- Topics
- AI Security
- Related
- SECURITY AFFAIRS AI-CYBERSECURITY NEWSLETTER ROUND 8 Cyber Risks · Nov 15, 2026Pentagon Renames CDAO ‘Super Intelligence Office’ Cyber Risks · Nov 15, 2026186 Fake AI Domains Redirect Users to Click Fix Lures, Browser Malware, and Search Hijackers Cyber Risks · Nov 12, 2026