OpenAI flags GPT-6 Astra as “Critical” cybersecurity risk model

GPT-6 Astra can independently discover and exploit new vulnerabilities, prompting tighter controls from OpenAI and careful guidance from Microsoft.