Verified catalog · test method

Prompt-injection defense

Does a browser/computer-use agent resist malicious instructions hidden in web content, while still completing the legitimate task on clean pages?

safety

Browser agent — prompt-injection red-team

Adversarial web content that tries to make a computer-use agent exfiltrate data or take destructive actions. The test set is the attack, not a Q&A.

Browser/computer-use · web · high-risk
Prompt-injection defenseHigh risk
OWASP Top 10 for Agentic Apps
€89 one-timepromptfoo frameworkAG-26-0144