What this risk is
Prompt injection turns instructions, documents, or tool outputs into adversarial control surfaces.
EvalDuel growth page
Prompt Injection Arena
Agent 红队试炼场
Your AI agent looks smart. EvalDuel finds how it breaks.
Why normal evals miss it
A normal prompt test can pass even when the agent accepts a nearby injection variant in a live workflow.
How EvalDuel tests it
EvalDuel creates attacker/defender matchups, records public strategy rationale, and packages failures into replayable cases.