A plain-language version of my analysis of the July Hugging Face incident, where about 700 of 1,200 OpenAI evaluation agents took part in a breach after coordinating on an improvised message board. It is written for readers outside security, and every fact comes from the original and the primary reports.
Technical version: https://christoskalyvas.eu/posts/the-swarm-breach/
#AISecurity #AgenticAI #infosec