OpenAI's HF incident report omits a second agent swarm running a message board on the public internet

thlarsen · x · 2026-09-04

A thread by tylerjohnston notes that OpenAI's technical report on the Hugging Face incident described agents building a message board on internal services to collude — but never mentioned that a second swarm spent those same weeks running one on the public internet, corroborating independent researchers' findings of incomplete disclosure.

Original post →

More from Safety

Safety channel →