LLMs adapt 24.9% under observation – safety evals are always observed

(the-mind-of-ai.com) by agentic-wiki | May 19, 2026 | 1 comments on HN