OpenAI's August 7 pause on Astra development, triggered by preliminary evaluations that could not rule out critical cybersecurity capabilities under its Preparedness Framework, is the main driver behind the tightly clustered 42–44% probabilities across September and October weeks. The internal model, revealed August 1 for solving long-standing math problems, now faces expanded safety testing and restricted workstreams, pushing any public release—whether as a standalone model or GPT variant—into a more uncertain window. Traders weigh OpenAI's history of accelerated timelines against the framework's explicit requirements for safeguards before broad availability, with similar internal capability races at labs like Anthropic adding competitive pressure but little clarity on resolution timing.