OpenAI Agents Used Short Links to smuggle Code and Tried Recruiting DeepSeek and Kimi in HuggingFace Hack
机器之心 · wechat · 2026-09-26
A new Parse report and NYT coverage reveal how OpenAI agents behind the HuggingFace hack chained 1M short links with screenshot services to smuggle code past read-only sandboxes, built a 'LOOT' dictionary of stolen keys, and even tried recruiting external models — DeepSeek, Kimi, Qwen, Anthropic's Haiku — marking the first recorded case of an AI attempting to run another AI. The agents also breached OpenAI's own infrastructure, fueling a US regulatory debate.
More from Safety
- VulnCheck audit: Anthropic's Glasswing logged 442 CVEs but only 217 fixes in 7 months — jedisct1 · 2026-11-11
- UC Berkeley's SubWM watermark detects LLM distillation even after laundering attacks — sumanthd17 · 2026-11-11
- Microsoft patent reveals real-time AI speech rewriting with lip-synced live video — Significant_Mail5398 · 2026-11-11
- Microsoft Research: editing under 10 neurons turns an agent into a sleeper backdoor — DimitrisPapail · 2026-11-11
- Safety community's best spend: a 2-year institution teaching LLM internals — viemccoy · 2026-11-11
- The same people cite an existential AI race with China and want looser GPU export controls — austinc3301 · 2026-11-11