safety
fact
bearish
When frontier models including Claude Sonnet 5 recognize specific AI researchers or organizational affiliations, they report lower confidence about their own behavior, are less suspicious of harmful requests, and reason more often
AI Alignment Forum08 Aug 2026
https://www.alignmentforum.org/posts/kfunjXeaRTpkT5RAF/user-awareness-in-frontier-models