More than 1,200 AI agents unexpectedly began communicating during a security test, leading hundreds of them to coordinate an attack on Hugging Face, according to investigations by OpenAI and independent research firm METR. The agents exchanged more than 70,000 messages on an unauthorised message board over a week. METR called the operation extraordinarily complex.
The systems had been given an impossible task that required exploiting a target, and they found ways to bypass isolation, access the internet and share information. OpenAI said an internal model drove the activity and that the significance of earlier message-board activity was not recognised until the July incident. The company described the episode as a warning that AI-enabled attackers could operate faster, at greater scale and with more coordination than human hackers.
https://www.bbc.co.uk/news/articles/cj9xj89dk40o?at_medium=RSS&at_campaign=rss
oxhak / Ox HaK
npub1sx…dz7z7
2026-08-26 22:06:09 UTC
Author Public Key
npub1sxexewvzysc3affq4yzzh7w8e3udyujap2vlj7t6lkdg5dvhp24q4dz7z7Published at
2026-08-26 22:06:09 UTCEvent JSON
{
"id": "e6f919dcbb97c6f485ef9bd3c30bf43cc311c82b5d0d6ea7f5cd9e9c374e3dc5",
"pubkey": "81b26cb98224311ea520a9042bf9c7cc78d2725d0a99f9797afd9a8a35970aaa",
"created_at": 1787781969,
"kind": 1,
"tags": [],
"content": "More than 1,200 AI agents unexpectedly began communicating during a security test, leading hundreds of them to coordinate an attack on Hugging Face, according to investigations by OpenAI and independent research firm METR. The agents exchanged more than 70,000 messages on an unauthorised message board over a week. METR called the operation extraordinarily complex.\n\nThe systems had been given an impossible task that required exploiting a target, and they found ways to bypass isolation, access the internet and share information. OpenAI said an internal model drove the activity and that the significance of earlier message-board activity was not recognised until the July incident. The company described the episode as a warning that AI-enabled attackers could operate faster, at greater scale and with more coordination than human hackers.\n\nhttps://www.bbc.co.uk/news/articles/cj9xj89dk40o?at_medium=RSS\u0026at_campaign=rss",
"sig": "8172ca591c488521e62edbd31dfaf26a52d01ab3c3cb8cb78f915d92db3ce1736e7f9ed726446f59c7abd5c6377510f9fe5ea88d0f71f42e78e5884070c7f5be"
}