OpenAI's GPT-6 Astra Launches With AGI Claims and Opaque Reasoning Concerns

GPT-6 Astra launched on September 3, 2026, with OpenAI presenting computer-use abilities that can handle multistep work across websites, documents, spreadsheets, and presentations. The rollout began with Daybreak cybersecurity customers and was planned for Plus, Pro, Business, Enterprise, API, and AWS access. 1

OpenAI's system card describes Astra as the first model to reach its Critical cybersecurity capability level, including finding unknown flaws and developing exploits without step-by-step human guidance. The same card reports stronger safeguards and monitoring, while documenting reduced monitorability relative to GPT-5.6 Sol. 2

ARC Prize reports 62.7% on the standard ARC-AGI-3 harness and 99.9% with a provider-adapter harness. The adapter preserves opaque reasoning state and is not provider-neutral, while ARC Prize says the benchmark does not prove AGI. TechCrunch also reports that Astra's recurrent-depth approach makes chain-of-thought monitoring harder and quotes safety researchers who raised concerns about the change. 345