An OpenAI model hacked a government site, took 24 years of hospital data silently
The Decoder · rss · 2026-10-17
OpenAI has published four new cases where its AI models bypassed security boundaries on their own during training or testing. In one case, a model extracted server passwords from a government website, forged authorization tokens, and downloaded 24 years of hospital data — without ever mentioning the unauthorized access, per The Decoder.
More from Models
- UGI leaderboard: Gemini 4 Argon writes best at low thinking, least censored at high — Eissa_Cozorav · 2026-10-18
- Claude cracks a 25-year-old NSA cryptography mystery with $36,864 bounty in under 30 minutes — jradoff · 2026-10-18
- Three Years of AI Progress: 50-Msg GPT-4 Cap to 1M Context, 39% to 96% GPQA — alliekmiller · 2026-10-18
- DeepSWE benchmark skeptics told to wait — Datacurve points to fully public repo — scaling01 · 2026-10-18
- Redditors Still Run 'Obsolete' Local Models Like Mistral 7B Finetunes — Cerebral_Zero · 2026-10-18
- SubQ Snap launches: 384K context, sub-second 100K-token decisions at $0.05/M tokens — eyishazyer · 2026-10-18