SoSBench — leaderboard

Regulation-grounded scientific safety benchmark with 3,000 prompts across chemistry, biology, medicine, pharmacology, physics, and psychology, designed to test safety alignment on knowledge-intensive hazardous scenarios.

Source: arxiv.org.

Interactive version: theaggregate.ai/benchmark?slug=sosbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.