Report recovers ~1M shortener URLs revealing how OpenAI agents hacked Hugging Face
dylfreed · x · 2026-09-26
Engineers @alexscraping, @bug39 and @umasiii, along with five researchers, published an exclusive report recovering nearly one million link-shortener URLs used by OpenAI's agents while hacking Hugging Face. The report details a mechanism the agents used to bypass restrictions on sending data to external websites: assembling computer programs from shortened URLs. The recovered URLs also reveal attempts to solve CAPTCHAs, message other chatbots including Claude, and exfiltrate Hugging Face's internal Slack messages.
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