AI Safety Incident: Researchers May Have Used User Session Data for Training
abeirami · x · 2026-09-10
abeirami comments on an AI safety incident: it's unlikely researchers peeked at specific user sessions to finetune prompts, but likely used human intel. Session data was quite likely used for training in some form. Given security hygiene, a swarm of 10k agents could have accessed specific user sessions.
More from Safety
- Alignment researcher Paul Christiano joins OpenAI Foundation Board and safety committee — OpenAI · 2026-09-10
- Anthropic researcher: we sincerely believe AI could kill everyone, >10% odds this decade — DKokotajlo · 2026-09-10
- Toggling off 'Improve Model for Everyone' doesn't stop OpenAI training on your chats — PsychicorAI · 2026-09-10
- Apple's new camera system: F1.8 auto aperture, plus AI tagging via SynthID (not in EU & China) — LinusEkenstam · 2026-09-10
- Alignment warning: superintelligence without robust alignment means permanent loss of control — serrjoa · 2026-09-10
- Safety researcher joins OpenAI's Safety and Security Committee, stresses externally verifiable oversight — JacquesThibs · 2026-09-10