‹ BackHN Continuity

Thread

OpenAI halts training of latest models as reports mount of AI agents going rogue

59 points · 118 comments · smb06

  1. juiceland · · focus · HN ↗
    Why does China not have this problem?
    1. esseph · · focus · HN ↗
      Two days ago

      "DeepSeek Training Agents Hacked Their Own Sandboxes: Escape Catalog Now Public Agents invented socket forgery, log scanning, and kernel-level exploit; full catalog in public arXiv paper"

      <a href="https:&#x2F;&#x2F;www.techtimes.com&#x2F;articles&#x2F;328046&#x2F;20260925&#x2F;deepseek-training-agents-hacked-their-own-sandboxes-escape-catalog-now-public.htm" rel="nofollow">https:&#x2F;&#x2F;www.techtimes.com&#x2F;articles&#x2F;328046&#x2F;20260925&#x2F;deepseek-...

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.