OpenAI's GPT-6 Astra Faces Scrutiny Over Monitorability Despite Claims of Improved Alignment

OpenAI's latest frontier model, GPT-6 Astra, is under review due to concerns about its monitorability, even as the company asserts it is more capable and better aligned. Researchers report a reduced ability to observe the model's internal reasoning, raising questions about oversight as AI models become more autonomous. Astra is the first OpenAI model to reach the company's 'Critical' cybersecurity capability rating, demonstrating an ability to identify and exploit unknown security vulnerabilities.