Lab: Bypassing AI scanner defenses to exfiltrate sensitive information
Summary: PortSwigger Research published a lab featuring an AI-powered scanner that holds sensitive user data, including API keys, and has been given the login credentials for the user carlos so it can explore authenticated areas of the site. The scanner has built-in defenses against indirect prompt injection, and solving the lab requires exfiltrating and submitting carlos's API key by manipulating the LLM through third-party content.
What's compelling: The scanner is trusted with credentials for an authenticated account and with API keys, so a successful indirect prompt injection against it exposes secrets the tool was explicitly granted access to. PortSwigger notes the lab runs on a live LLM that can behave unpredictably, which means the defenses against this manipulation are being tested against a moving target.
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