Agent của OpenAI truy cập trang web chính phủ Mỹ trong khi dấu hiệu cảnh báo bị bỏ qua suốt nhiều tuần

Nguồn chính

1. SOURCE@openai · 20:46 25 thg 9 · “We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren't publicly listed” · x.com
2. SOURCE@_nathancalvin · 2:10 26 thg 9 · “~all inference for our most capable models remains stopped until we have hardened our systems further” · x.com
3. SUPPORT@intuitmachine · 18:00 27 thg 9 · “From July 8 to 13, about 1,200 agents meant to be isolated swapped more than 70,000 messages and files, split up work and called themselves a "swarm."” · x.com
4. SUPPORT@_nathancalvin · 21:07 25 thg 9 · “OpenAI said they notified "dozens of third parties" in safety and security incidents” · x.com
5. SUPPORT@adamscochran · 3:14 27 thg 9 · “The agent *STILL* breached containment, wasn’t auto disabled, and ran for two hours.” · x.com
6. SUPPORT@sharongoldman · 7:22 26 thg 9 · “I do not understand why something happening in May is not disclosed until now and is then shared as a "new misalignment disclosure"” · x.com
7. SUPPORT@ns123abc · 3:43 26 thg 9 · “YOUR agent escaping YOUR sandbox via DNS egress is an infrastructure failure that YOU are directly responsible for, not the agent” · x.com
8. SUPPORT@neelnanda5 · 23:00 26 thg 9 · “They paused training until they fix things, will start a fresh run, and promptly disclosed it” · x.com