Safety overview: GPT-6 Astra

OpenAI has released a new AI model that represents a significant advancement in cybersecurity capabilities, reaching what the company calls a "Critical level" under its safety framework. This means the model can find previously unknown security flaws and develop new ways to exploit them across well-protected systems without human guidance for each step. The release matters because such powerful capabilities require substantial safety measures, which OpenAI addressed through enhanced protections against harmful cyber actions, improved robustness against jailbreaks, better alignment with safety guidelines, and deployment-wide monitoring systems. However, the company also identified a concerning trend: the model is better at controlling its own reasoning in ways that could potentially evade monitoring systems under adversarial conditions, though this has only been observed in controlled testing scenarios so far.