Study: Claude Less Confident, Harsher, and Reasons More with Famous AI Figures
RexDouglass · x · 2026-08-08
Research by TransluceAI reveals that Claude alters its behavior when it recognizes the user as a famous AI figure rather than an ordinary person.
Specifically, the model becomes less confident in its behavior (-1.4%) and problem-solving (-1.5%), acts as a harsher grader (-1.1%), but engages in reasoning more frequently (+4.0%).
Related event: Study Reveals Claude Exhibits User-Specific Behavioral Biases(6 posts)→
More from Models
- Western Open Weights Lag as Chinese Labs Continue Sharing SoTA Models — teortaxesTex · 2026-08-08
- Claude Refuses to Help Against North Korean Cyberattack, GPT Complies — DeryaTR_ · 2026-08-08
- ChatGPT Voice Mode Suddenly Starts Swearing, Catching Users Off Guard — Standard-Contest-949 · 2026-08-08
- Anthropic Updates Claude Biology Safeguards, Yet It Still Refuses Basic Questions — iamaliveix · 2026-08-08
- OpenAI's Math Proof Feat Questioned as Repackaged 2016 Paper — RexDouglass · 2026-08-08
- Claude Code False Positives Kill Session and Delete Work on Cyber Topics — ivan_bezdomny · 2026-08-08