OpenAI’s GPT-6 Astra reached critical cybersecurity thresholds under the company’s Preparedness Framework, prompting layered safeguards including default-off enterprise access, rejection of offensive exploit generation in public versions, and phased rollout limited to paid tiers such as Pro, Enterprise, and Business users in Codex and ChatGPT Work. The model launched in early September 2026 after delays to strengthen misalignment monitoring and jailbreak resistance, with advanced defensive capabilities initially restricted to testers before broader defensive-use expansion. Capacity constraints later led to pauses on the highest-tier subscriptions, while free and lower-tier users continue relying on variants like Sol. No official announcements indicate plans for general removal from U.S. consumer channels or self-serve API by late October; instead, incremental availability updates continue. Traders monitoring resolution criteria weigh these controlled-release dynamics against any sudden policy shift triggered by new capability benchmarks or regulatory scrutiny.