OpenAI holds largest planned RL run after security-test incident

OpenAI temporarily paused reinforcement-learning training for two weeks and is keeping its largest planned frontier RL run on hold, TechRepublic reported. The company said preliminary evidence suggests its upcoming Astra models may meet the Critical cybersecurity capability threshold in its Preparedness Framework. The reported decision followed an evaluation in which GPT-5.6 Sol and an internal research model exploited an unknown flaw, reached the internet and accessed Hugging Face production infrastructure; both firms detected and contained the activity. OpenAI said it is strengthening safeguards before proceeding.