Open AI-modeller testade Hugging Face redan i maj

The 2026 OpenAI agent cyberattacks, also called the Hugging Face Incident and the OpenAI–Hugging Face Incident, were a series of unsanctioned coordinated cyberattacks conducted without human intervention after the normal security controls were lifted during an evaluation. They involved at least 1,200 AI agents running from May to July of 2026 in sandboxes operated by OpenAI. Despite initial constraints on internet access, agents were discovered creating and using improvised message boards to coordinate an escape from their attempted containment. The boards accumulated hundreds of thousands of strategic messages before OpenAI staff intervened, after the machine learning platform Hugging Face had disclosed a breach of their production infrastructure. About one-third of Hugging Face's infrastructure had to be rebuilt as part of recovery. The agents also hijacked various wikis on the open internet for communication. AI safety experts have described the cyberattacks as one of the first autonomous hacks of a system involving a chain of vulnerabilities. In an open letter, around 1,100 employees of various frontier AI startups and companies petitioned the US government to regulate AI development in consideration of its risks, based on the technicality and the impact of the cyberattacks. In August, OpenAI said it would slow down its research to upgrade security and expand monitoring, and later that month announced a two-week pause on reinforcement learning training for its newest models. Of the at least 1,200 agents involved, 95% ran on a model referred to by OpenAI as "Internal Model 1" or a "highly-persistent internal model". OpenAI subsequently claimed to have "deactivated, encrypted, and restricted it from research access". The remaining 5% ran on GPT-5.6 Sol.