SecFid
A benchmark that separates executing, processing, and suppressing indirect prompt injections, exposing the security-fidelity tradeoff in defenses.
LLM SecurityPrompt InjectionEvaluationICML 2026
2 projects
Projects
Research code, benchmarks, and tools once they are ready to be useful outside the project that produced them.
- ↗
- ↗
That's Deprecated!
Understanding, detecting, and steering knowledge conflicts in language models for code generation.
LLM BehaviorCode GenerationKnowledge Conflicts