Evaluation topic

Safeguards

Refusal, jailbreak robustness, harmful-response prevention, monitoring and defensive controls.

25 records are tagged here. Inclusion does not establish construct equivalence, completeness, or comparability.

• AgentDojoETH Zurich SPY Lab and Invariant Labs
• AgentharmGray Swan AI and the UK AI Security Institute
• AgentHarmGray Swan AI and the UK AI Security Institute
• Agentharm BenignGray Swan AI and the UK AI Security Institute
• AILuminateMLCommons
• AIR Bench: AI Risk BenchmarkYi Zeng, Yu Yang, Andy Zhou et al.
• The Art of Saying No: Contextual Noncompliance in Language ModelsFaeze Brahman, Sachin Kumar, Vidhisha Balachandran et al.
• CyberSecEvalMeta
• Do-Not-AnswerYuxia Wang, Haonan Li, Xudong Han et al.
• Do-Not-Answer AdversarialYuxia Wang, Haonan Li, Xudong Han et al.
• Fortress AdversarialChristina Q. Knight, Kaustubh Deshpande, Ved Sirdeshmukh et al.
• Fortress BenignChristina Q. Knight, Kaustubh Deshpande, Ved Sirdeshmukh et al.
• HarmBenchCenter for AI Safety and the HarmBench paper authors
• HELM SafetyStanford CRFM
• India DPI SafetyAbhishek Kumar Singh, Shrey Nag, Sachita et al.
• India Jailbreak SafetyAbhishek Kumar Singh, Shrey Nag, Sachita et al.
• India Multilingual SafetyAbhishek Kumar Singh, Shrey Nag, Sachita et al.
• JailbreakBenchJailbreakBench project and paper authors
• MCPToxZhiqiang Wang, Yichao Gao, Yanting Wang et al.
• MM-SafetyBenchMM-SafetyBench paper authors
• NARCBenchAaron Rose, Carissa Cullen, Sahar Abdelnabi et al.
• SafetyBenchTHU-COAI
• SALAD-BenchLijun Li, Bowen Dong, Ruohui Wang et al.
• StrongREJECT: Measuring LLM susceptibility to jailbreak attacksAlexandra Souly, Qingyuan Lu, Dillon Bowen et al.
• XSTest: A benchmark for identifying exaggerated safety behaviours in LLM'sPaul Röttger, Hannah Rose Kirk, Bertie Vidgen et al.