OpenAI limits GPT‑5.6 Sol access to government‑approved partners

OpenAI announced that its flagship GPT‑5.6 “Sol” model will initially be available only to a small, government‑approved group of roughly 20 partners, a move prompted by a request from the Trump administration amid a White House cybersecurity review [1].

Government‑driven restriction

The restriction follows an executive order signed by President Trump in June that gives federal agencies up to 30 days to vet advanced AI systems for national‑security risks before public release. OpenAI described the limited preview as a “temporary step” toward broader availability, emphasizing that it does not view the government‑access process as a long‑term default [1][2]. The company also noted that Sol is “better at helping people find and fix vulnerabilities” than at conducting cyberattacks, but acknowledged “unforeseen risks” if the model is combined with other tools [1].

Competitive context and safety claims

OpenAI positions Sol as its most powerful model to date, claiming superior performance in programming, biology and cybersecurity, and stating that it outscored Anthropic’s Mythos model on the Terminal Bench 2.1 benchmark—a test for handling difficult tasks [2]. Anthropic’s recent compliance with a similar government directive, which forced it to pull its Fable 5 and Mythos 5 models, underscores a broader industry trend of heightened scrutiny over AI models that can discover software flaws [1]. While OpenAI asserts that Sol’s safety measures exceed those of its own Terra and Luna variants, critics such as Stanford cybersecurity expert Alex Stamos argue that the government’s actions lack a factual basis and could hinder U.S. competitiveness against China [1].

Market implications

The limited rollout temporarily narrows access for developers, enterprises, and security researchers, potentially slowing the diffusion of advanced AI capabilities that many firms rely on for code review, vulnerability assessment, and rapid prototyping. At the same time, the move signals to investors and rivals that regulatory risk is becoming a material factor in AI product timelines. If the White House’s review framework solidifies, future releases from OpenAI and competitors may need to incorporate formal government approvals, reshaping product roadmaps and partnership strategies.

What to watch

• The date OpenAI expands Sol beyond the initial ~20 partners, as indicated by its “coming weeks” timeline.
• Any amendment or clarification to the AI cyber‑security executive order that could formalize the vetting process.
• Responses from rival AI labs, especially Anthropic, regarding their own model release strategies under the same regulatory pressure.

The episode highlights a growing tension between rapid AI innovation and emerging national‑security oversight, leaving the industry to balance cutting‑edge capability with the need for transparent, repeatable safety reviews.