WASP

Capture Summary

Benchmark for web agent security against prompt injection attacks. The abstract emphasizes realistic web agent hijacking objectives and an isolated environment for testing attacks against web navigation agents.

Relevance

• Important benchmark source for measuring prompt injection risk in web agents.
• Provides a more realistic evaluation framing than isolated single-goal attacks.
• Useful for research questions on measurable agent compromise and realistic adversary constraints.

Collection Notes

Collected as a benchmark source for Security for AI evaluation work.