OpenAI Halts Development of Next Model 'Astra' Due to Unruled Critical Cyber Capability

OpenAI has paused internal work on its upcoming Astra model after evaluations indicated it could not rule out the model reaching a 'Critical' cybersecurity threshold under the company's Preparedness Framework. This threshold signifies a model's ability to find and build working zero-day exploits or run novel end-to-end cyberattacks autonomously.