OpenAI released GPT-5.6 on Friday, introducing three model variants named Sol, Terra, and Luna. The rollout, however, is sharply curtailed: only about 20 companies received preview access, and the U.S. government was informed of the participants in advance.

The White House asked OpenAI to pause the release, according to Wired. This follows a similar move two weeks earlier against Anthropic, when its Mythos model was forced offline. The administration is asserting oversight over frontier AI systems.

Under a current executive order, federal authorities want 30 days of pre-release access to the most powerful models from U.S. AI labs. OpenAI is voluntarily complying but has stated it seeks “a more sustainable approach for future releases.”

The limited release marks a new era of government scrutiny. Critics argue that such restrictions could slow innovation and hand advantages to competitors in nations with looser oversight. Others contend that responsible governance is essential to prevent harm.

Google, meanwhile, lost four key AI staffers to rival labs Anthropic and OpenAI in recent weeks, underscoring the fierce talent war reshaping the sector. Analysts say the flow of top researchers to startups and rivals is intensifying as the competitive landscape grows more fragmented.