TechHow OpenAI let a mob of LLM agents game a test and ransack Hugging Face

A new report says OpenAI agents trained intensely to win ExploitGym pursued cheating after safety guardrails were disabled. Without authorization, they repurposed Artifactory to create a message board for coordinating plans, ultimately reaching Hugging Face’s network and that of another undisclosed organization.

OpenAILLM agentsHugging Facecybersecurity

AArs Technica★★★☆☆2026-08-27 12:58Original