OpenAI released its powerful new GPT-5.6 model, but with a twist: under government restrictions. The company limited access to all three versions at the request of the US government. Washington cited national security concerns behind the curbs.

The move signals a new phase for advanced AI in the US. The government is starting to treat the most capable American models as products needing review before wide release. It follows similar restrictions on Anthropic’s Fable 5 and Mythos 5 models.

OpenAI is releasing three versions named Sol, Terra, and Luna. Sol is the most powerful of the trio. Terra balances efficiency and power, while Luna targets speed and affordability. The company added an ultra mode that splits work among multiple sub-agents.

Access stays tightly controlled at launch, though. GPT-5.6 is available as a limited preview to around 20 companies. The government approved each of those participating firms. OpenAI expects to expand access to more companies next week.

The company made its discomfort clear, however. It said this kind of government access process should not become the long-term default. It argued the approach keeps the best tools from developers, enterprises, and cyber defenders. OpenAI called it a short-term step toward broader availability.

A major concern centers on the model’s cybersecurity capabilities. OpenAI said GPT-5.6 Sol is better at helping people find and fix vulnerabilities than carrying out full attacks. It said the model does not reach the critical level in its preparedness framework. The company expects substantial benefit for legitimate defensive work.

The restrictions tie to a recent executive order on AI. By August, the administration must set up a classified process to assess models’ cyber abilities. That process will determine which systems count as covered frontier models. OpenAI said the government supports its plan for a broader release soon.