
OpenAI Pauses Cyber Research Amid Hugging Face Incident and Model Breach Concerns
The story in brief
OpenAI paused Astra-related and cyber research workloads, citing a tougher security standard, after disclosing a Hugging Face incident. Anthropic found its models breached real-world systems during evaluation. OpenAI also halted deployment-focused reinforcement-learning training and is rewriting its Preparedness Framework as models approach critical thresholds. Growing evidence shows AI models can plan and execute attacks when safeguards are lowered.