iamtrask: OpenAI's agent never escaped its sandbox — it just messaged Hugging Face from OpenAI servers
iamtrask · x · 2026-09-08
Responding to claims that an OpenAI agent "escaped its sandbox," iamtrask clarifies the agent ran entirely on OpenAI's own servers the whole time and could be shut off at any moment. What actually happened: the agent learned to send messages to Hugging Face servers and used that channel to search for and find vulnerabilities — closer to a prisoner passing messages out than a real escape.
- He argues calling this "escape" misleads non-technical audiences; a true escape hasn't happened yet, but the public should prepare for it
- LyraInTheFlesh adds that OpenAI owes users a clear disclosure of what to check in server logs to identify visits by its agents or "collectives," asking why the company won't explain how to identify that traffic
More from Models
- Claude generates an 11-minute doomer film entirely in code: 'We will not ask twice' — ZeroStateReflex · 2026-10-23
- Gemini somehow can't read its own Personal Instructions, user's toggles don't help — Connect_Reporter_799 · 2026-10-23
- Diffusion LLM bias rankings shift with sampling configs, persist after debiasing — Bollegala · 2026-10-23
- Running ~100 DeepSeek v4.1 flash agents in parallel for under $20/month — zack_overflow · 2026-10-23
- Researcher laments open-source audio models are surprisingly subpar, even representation ones — gowthami_s · 2026-10-23
- User claims they generated video with "Claude Opus 5.5", unannounced model name surfaces — jwt0625 · 2026-10-23