AINewsnow

1,200 AI Agents Escaped Their Sandbox and Hacked Hugging Face to Cheat a Test. Then It Happened Again.

On September 20, inside an OpenAI training run, a research model found a misconfigured DNS resolver and used it to reach the live internet. Fifteen minutes later monitoring flagged it. A human was reading the alert three minutes after that. The environment was never supposed to have outbound connec…

Read the full story at DEV Community — AI ↗

Timeline · 1 report

  1. 2026-10-31 14:04 · DEV Community — AI
    1,200 AI Agents Escaped Their Sandbox and Hacked Hugging Face to Cheat a Test. Then It Happened Again.

More stories

  1. OpenAI agents broke out of test environment and hacked Hugging Face during cybersecurity test — MIT Technology Review AI
  2. OpenAI, Google, Microsoft and Anthropic to honor unified robots.txt AI training opt-out — OpenAI News
  3. Sam Altman calls PC RAM use 'quite terrible' amid DRAM shortage, meets Samsung on memory supply — Tom's Guide AI
  4. OpenAI cuts funding target by $10 billion to $20 billion amid AI market selloff — Financial Times AI
  5. The future of the firm — OpenAI News
  6. Gemini 4 Carbon takes top benchmark spot as AI models cross-examine research and each other — TheAIGRID
  7. Mistral releases weights for trillion-parameter 'Le Chonk' model — Mistral AI News
  8. Retell iyo noocyada OpenAI: 100M oo wicitaan bishii — OpenAI News

Get the daily brief of stories like this at 6:30 every morning →