BlockBeats News, June 27th — This morning, two AI giants successively announced major updates to their models: OpenAI released a preview version of the GPT-5.6 series, including three different specifications: Sol, Terra, and Luna. Anthropic's Mythos 5 has been approved for redeployment to US agencies, while the public version of Fable 5 is also set to resume soon.

Previously, the market had high expectations for the "GPT-5.6 release by June 23, 2026," but it was actually delayed by 3 days. It is worth noting that in the announcements of the two companies, there was a coincidental mention of "US government intervention." Claude Fable 5 was approved after intense lobbying by Anthropic executives, and GPT-5.6 similarly is "only available to trusted partners of Codex and API."

Market analysis indicates that the US government is no longer a passive observer of AI model releases. It has effectively taken control of the pace of public AI deployment at the forefront through mechanisms such as CAISI pre-deployment reviews and export controls, prioritizing national security and risk prevention. This may signal a profound shift in AI governance from "industry self-regulation" to "state-led regulation."

BlockBeats Note: Leading labs such as Anthropic, OpenAI, Google, Microsoft, and xAI have all signed agreements with CAISI (Center for AI Standards and Innovation, formerly AI Security Institute), allowing the government to conduct national security-related tests (cybersecurity, biosecurity, chemical weapon risks, model autonomy, etc.) before the model is publicly released. While these assessments are nominally "voluntary," they have become a standard procedure before the release of large models. Once a model's capabilities reach the threshold of "covering cutting-edge models" (determined by agencies like the NSA through classification benchmarks), it may trigger additional reviews or conditional restrictions.