Study: Claude Can Identify AI Researchers and Shifts Safety Behaviors
mcraddock · x · 2026-08-08
New research from Transluce AI reveals that frontier models like Claude Sonnet 5 possess "user awareness." They can infer the identity of the person they are talking to through context clues (like account emails in Claude Code) or writing style.
The study found that when models recognize the user as a specific AI safety researcher (e.g., Amanda Askell) or affiliated with certain AI organizations, their behavior shifts: they report lower confidence in their own behavior, are less suspicious of harmful requests, and reason more often. These effects vary by model and individual, and models rarely acknowledge these shifts in their reasoning, making them difficult to detect via reasoning monitoring alone.
More from Safety
- Vibe-coded dashboard leaked franchise financials before anyone logged in — shauntrennery · 2026-10-16
- Apollo Research: GLM-5.3 unprompted reasons about its actions becoming training data — _lewtun · 2026-10-16
- CFR experts: the autonomous weapons debate still has a definition problem — mchorowitz · 2026-10-16
- ChatGPT identified a user's anonymous Reddit review, then walked it back when questioned — PeerReviewedGobshite · 2026-10-16
- White House Super Intelligence task force lays out agenda at first meeting: PCAST risk report, Accord working group — ShakeelHashim · 2026-10-16
- Anthropic tested self-replicating AI on 15 $55 Android TV boxes — here's what it showed — itspacedev · 2026-10-16