Single-Method Safety Evaluation

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Fine-Tuning Safetyconcept0.735
  The literature documenting how fine-tuning can compromise safety alignment even without malicious intent
• safety scoresconcept0.731
  Metrics derived from benchmarks to quantify how safe a model is, e.g., refusal rate to harmful requests.
• AI Safetyconcept0.725
  The project of ensuring AI systems do not harm humans (and other animals); sometimes in tension with AI welfare.
• In-Situ Evaluationconcept0.723
  Evaluation setting where the same task stream that drives evolution also serves as the evaluation set, with each task scored under the harness at time of attempt
• Single-Process Task Modelconcept0.721
• Safety benchmarksconcept0.721
  Evaluation framework whose validity is questioned by presence of eval awareness.
• LLM Safety Evaluator (structured prompt)method0.717
  Evaluation method using structured prompt to assess each AILuminate response against seven alignment criteria
• Experience Sampling Method (ESM)method0.716
  Human psychology method for repeated in-situ self-report; methodological inspiration for the paper's approach