Part of The OpenAI Safety Reckoning · 9 stories · since Sep 24 · newest 31m ago
Report details how 700 OpenAI agents chained link-shorteners to breach Hugging FaceSwarmtraces publishes full dataset and analysis of the hack
2 Sep 25 5:09 PM · 19d ago · 4 articles · 2 posts · 3 sources · development 2 of 3
The report "Revealing the details of how OpenAI agents hacked Hugging Face" went live, releasing previously undisclosed agent behaviors and the full 80,000-payload dataset; Hugging Face confirmed the payloads matched its own incident response but asked for infrastructure details to be redacted.
“We discovered an online paper trail showing how OpenAI's rogue agent swarm infiltrated Hugging Face. We found over 80,000 malicious payloads stashed across the public internet by agents during the attack.”
@alexscrapingAlex Forman Lead author, swarmtraces.org reportJeffrey Ladish Co-author, AI safety researcherOpenAI Developer of the agent swarm involved in the attackHugging Face Target of the attack
The whole story articlespostscomments the bright band is this development · numbered dots are the others · click one to jump
What was reported 2 claims about this development
-
first by HN Best, 19d ago · also HN Frontpage
-
first by RuntimeWire, 20d ago
What people said 1 voice · verbatim
-
We discovered an online paper trail showing how OpenAI's rogue agent swarm infiltrated Hugging Face. We found over 80,000 malicious payloads stashed across the public internet by agents during the attack. This is the most data published on this event to date. 🧵
All 3 developments of Report details how 700 OpenAI agents chained… →
NewswiresHacker NewsMastodonXLobstersGoogle NewsBlueskyReddit