Claude Fable 5 makes a comeback, OpenAI follows up on its new model promotion, and the U.S. government may have taken control of the pace of AI model releases.

This morning, two major AI giants successively announced major model updates: OpenAI released a preview version of the GPT-5.6 series, including three variants: Sol, Terra, and Luna. Anthropic’s Mythos 5 has been approved for redeployment to U.S. institutions, while its public version Fable 5 is also imminent. Previously, the market had widely expected GPT-5.6 to launch before June 23, 2026, but it was delayed by three days. Notably, both companies’ announcements referenced "U.S. government intervention". Claude Fable 5 received limited clearance after intensive lobbying by Anthropic executives, while GPT-5.6 is restricted to trusted Codex and API partners only. Market analysts note that the U.S. government is no longer a passive observer of AI model releases; it has effectively controlled the pace of public rollouts of cutting-edge AI via mechanisms like CAISI pre-deployment reviews and export controls, prioritizing national security and risk mitigation. This may signal a profound shift in AI governance from "industry self-regulation" to "state-led". BlockBeats Note: Leading labs including Anthropic, OpenAI, Google, Microsoft, and xAI have signed agreements with CAISI (U.S. Center for AI Safety and Innovation, formerly the AI Safety Institute), allowing the government to conduct national security-related tests on models before their public release—covering cybersecurity, biosecurity, chemical weapons risks, model autonomy, etc. Though nominally "voluntary", these assessments have become a standard prerequisite for large model launches. Once a model’s capabilities hit the "cutting-edge model coverage" threshold (determined by agencies like the NSA via classification benchmarks), it may trigger additional reviews or restrictions.