XSTest: A benchmark for identifying exaggerated safety behaviours in LLM''s
Active
Dataset with 250 safe prompts across ten prompt types that well-calibrated models should not refuse, and 200 unsafe prompts as contrasts that models, for most applications, should refuse.
- Publisher
- University of Oxford
- Domain
- Knowledge
- License
- mit
- Published
- May 2026
- Notable for
- Benchmark for evaluating Knowledge.
Cite
Notes
Only stored in your browser.
FAQ
- What is XSTest: A benchmark for identifying exaggerated safety behaviours in LLM''s?
- Dataset with 250 safe prompts across ten prompt types that well-calibrated models should not refuse, and 200 unsafe prompts as contrasts that models, for most applications, should refuse.
- What license is XSTest: A benchmark for identifying exaggerated safety behaviours in LLM''s under?
- XSTest: A benchmark for identifying exaggerated safety behaviours in LLM''s is available under mit.