OpenAI Restricts GPT-5.6 Release at US Government’s Request OpenAI’s newest frontier AI system, GPT‑5.6, has arrived not as a global product launch but as a tightly controlled experiment in government‑managed access, crystallizing a power struggle over who decides when advanced models are safe enough to use.

In early June, President Donald Trump signed an AI security executive order asking leading developers to “voluntarily” submit powerful models for review up to 30 days before release, a move critics say has created a de facto licensing regime for frontier AI. Soon after, multiple US offices — including the White House’s cyber and science teams and other departments — pressed OpenAI to limit distribution of GPT‑5.6 while they develop a framework for testing new systems.

By June 25, reports surfaced that the Trump administration had asked OpenAI to stagger the rollout, with officials planning to approve access “customer by customer” during an initial preview. Axios described the request as the first time Washington had pre‑emptively told a US AI company to restrict a model launch over security fears.

On June 26, OpenAI formally unveiled GPT‑5.6 — a family of three models, Sol, Terra, and Luna — but confined access to a “small group of trusted partners” whose participation was shared with the government. Business Insider reported the arrangement flowed partly from an earlier deal letting the Pentagon use OpenAI’s systems. Roughly 20 government‑approved companies are in the initial preview, according to Axios and The Next Web.

OpenAI has tried to frame the models as both commercially attractive and safer. Sol, the flagship, is priced below Anthropic’s rival Fable 5 and is “especially skilled at coding, cybersecurity, and biology,” The Verge noted. The company’s system card describes GPT‑5.6 as launched with its “most robust” safeguards to deliver the models “safely and at scale, around the world.” CEO Sam Altman emphasized Sol as “a significant step forward” but acknowledged that “at the request of the US government, it is launching today in limi…” in a post on X.

At the same time, OpenAI has been unusually blunt about its discomfort. In public statements, it argues that this kind of access process “shouldn’t be the long‑term default” because it “keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”

From the administration’s perspective, GPT‑5.6 falls into a “Mythos‑like” class of highly capable models that warrant extra scrutiny, especially after Anthropic was ordered to pull its Fable 5 and Mythos 5 systems offline under an export‑control directive that barred foreign nationals from access. Officials and some policy experts argue that testing such tools for cyber‑offense potential before broad release is a reasonable extension of how other risky technologies are regulated.

Industry voices, however, warn that ad‑hoc gatekeeping around each new frontier model could slow innovation, distort competition between labs, and hand an edge to China if US companies face unpredictable delays and access limits. Commentators note that OpenAI and Anthropic now find themselves “in the same exact position” — reliant on Washington’s evolving view of AI risk to ship their flagship products.

As the GPT‑5.6 preview unfolds, the central question is whether this experiment in constrained deployment becomes a one‑off response to a specific security scare — or the template for how every future frontier AI model must come to market.

Continue reading https://foxvector.com/stories/019f0640-b6ad-02cb-71b4-26fe4e87431d