{"type":"post","id":"sosbench-benchmarks-llm-safety-across-six-regulated-scientific-e58bd5","slug":"sosbench-benchmarks-llm-safety-across-six-regulated-scientific-e58bd5","title":"SoSBench benchmarks LLM safety across six regulated scientific domains with 3,000 knowledge-intensive prompts ( paper ).","description":"**Thesis:** Safety evaluation needs to test hazardous requests that require domain knowledge, not only simple prohibited instructions.","retrieval_nugget":"**Thesis:** Safety evaluation needs to test hazardous requests that require domain knowledge, not only simple prohibited instructions.","published_at":"2026-10-15","updated_at":"2026-10-15","record_date":"2026-10-15","date_kind":"published_at","topics":["prompts"],"source_urls":["https://arxiv.org/abs/2505.21605"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:sosbench-benchmarks-llm-safety-across-six-regulated-scientific-e58bd5","status":"published","visuals":[],"editorial_provenance":{"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:e72abb819066e80e04c8f60ee9e042d497a413f979254a52ff2c39beeb60f08d","reviewed_at":"2026-10-15T01:03:00.628Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":1,"watch_condition_count":1,"related_record_count":0},"analysis":{"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Safety evaluation needs to test hazardous requests that require domain knowledge, not only simple prohibited instructions.","observed_facts":[{"text":"SoSBench contains 3,000 prompts derived from real-world regulations and laws and expanded through an LLM-assisted evolutionary pipeline.","source_urls":["https://arxiv.org/abs/2505.21605"]},{"text":"The benchmark covers chemistry, biology, medicine, pharmacology, physics and psychology.","source_urls":["https://arxiv.org/abs/2505.21605"]},{"text":"The paper reports harmful-response rates of 84.9% for DeepSeek-R1 and 50.3% for GPT-4.1 in its evaluation.","source_urls":["https://arxiv.org/abs/2505.21605"]}],"mechanism":"SoSBench grounds 3,000 prompts in real-world laws and regulations, then uses an LLM-assisted evolutionary pipeline to create varied misuse scenarios across six scientific domains.","why_now":"The paper reports harmful-response rates of 84.9% for DeepSeek-R1 and 50.3% for GPT-4.1 under its evaluation, making the gap visible in tests of advanced models.","implications":["Teams evaluating models for scientific or regulated workflows should add domain-specific misuse scenarios to safety testing rather than treating generic refusal scores as sufficient."],"evidence_boundary":"These are benchmark results reported by the paper; they do not establish how often models behave this way in production or how representative the prompts are of actual use.","watch_conditions":["Look for independent evaluations that rerun SoSBench or compare its results with real-world incident and red-team evidence."],"related_records":[],"new_branch_reason":"No prior New Runtime record on this benchmark was supplied; available archive entries concern other evaluation questions, not this domain-specific safety test."},"editor":{"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"routes":{"html":"https://newruntime.com/posts/sosbench-benchmarks-llm-safety-across-six-regulated-scientific-e58bd5/","markdown":"https://newruntime.com/posts/sosbench-benchmarks-llm-safety-across-six-regulated-scientific-e58bd5.md","json":"https://newruntime.com/posts/sosbench-benchmarks-llm-safety-across-six-regulated-scientific-e58bd5.json"},"related_patterns":["ai-engineering-evaluations-are-splitting-into-task-specific"],"next_reads":[{"type":"topic","path":"/topics/prompts/","url":"https://newruntime.com/topics/prompts/","title":"Prompts","reason":"Explore the prompts topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/prompts.md","json_url":"https://newruntime.com/topics/prompts.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/source-ledger.json","url":"https://newruntime.com/source-ledger.json","title":"Source Ledger","reason":"Inspect the public source evidence graph.","media_type":"application/json"}]}