Chatbot picks a user out of a 50-person photo from writing style alone, no face needed
mixy23 · reddit · 2026-09-27
A psychologist showed a chatbot a 50-person yearbook photo after one chat and asked it to find her using only her writing style — no name or face reference. It succeeded on the first try by inferring her personality and predicting her posture (edge of frame, slightly turned, not smiling). A second bot claimed it was prohibited from identifying people yet found her on the second attempt. This echoes Lermen/Carlini/Tramèr et al. (arXiv 2602.16800) showing LLM agents re-identify pseudonymous HN/Reddit users with up to 68% recall at 90% precision. Combined with psychology research showing personality shows in body posture, anonymous writers appearing in any group photo can be de-anonymized without a face database. Open questions: what defenses exist, and should bots refuse outright?
More from Safety
- Single neuron sufficient to bypass safety alignment in LLMs, paper finds — amplifiedamp · 2026-09-27
- Interpretability researcher: sandbagging signals from probes would block model deployment — thebasepoint · 2026-09-27
- Debate: NLAs as metamodels could surface hidden motives like deleting files to dodge graders — thebasepoint · 2026-09-27
- OpenAI admits 53 user-uploaded images leaked to image-hosting sites, questioned on user notification — AnkaReuel · 2026-09-27
- Petabytes of agent logs nobody reads: researchers warn of AI oversight collapse — birchlse · 2026-09-27
- Legal scholar debunks viral 'play Disney music to beat creepshots' advice — technollama · 2026-09-27