OpenAI to Restrict GPT-5.6 Release at U.S. Government's Request

OpenAI to Restrict GPT-5.6 Release at U.S. Government’s Request OpenAI’s most powerful AI models yet — GPT-5.6 Sol, Terra, and Luna — are arriving under unprecedented political constraints, as the U.S. government asserts new authority over how “frontier” systems reach the public.

In early June, President Trump signed an AI security executive order that invites major AI firms to give the government up to 30 days of pre-release access to advanced models for testing, a framework critics say risks becoming a “de facto involuntary licensing regime.” Soon after, agencies including the Office of the National Cyber Director and the Office of Science and Technology Policy began previewing GPT‑5.6’s capabilities with OpenAI.

By June 25, reporting emerged that the Trump administration had formally asked OpenAI to “limit the release” of GPT‑5.6 to a small set of government‑approved partners over security concerns, the first time Washington had preemptively requested such a restriction from a U.S. AI company. Tech outlets described the White House as “asking OpenAI to slow roll the release” and to let the government approve access “customer by customer” during an initial preview period.

On June 26, OpenAI unveiled GPT‑5.6 in a “limited preview” focused on Sol, Terra, and Luna, touting Sol’s strength in coding, cybersecurity, and biology, and calling its protections “our most robust safety stack to date.” A parallel system card emphasized that the safeguards were “built to deliver these models safely and at scale, around the world.”

In practice, however, Sol initially went only to about 20 companies whose participation was individually cleared by the government, making this “the first time an American AI company has launched a frontier model under a government‑managed access list.” Business coverage noted that OpenAI’s Defense Department agreement was a key reason it complied, with the company stressing that “we don’t believe this kind of government access process should become the long-term default.”

Online, some AI leaders argued that tightly regulating big proprietary APIs while sparing open-source models is “quite rational” because “the most dangerous AI systems right now aren’t open models” but large frontier APIs. Others warned that ill-defined standards could slow U.S. deployments and hand an advantage to foreign competitors.

From Washington’s perspective, GPT‑5.6 — marketed as having “Mythos-like” cyber capabilities comparable to Anthropic’s recently restricted models — simply crossed a threshold where security vetting must come first. OpenAI, while cooperating, is betting that accepting short-term control will lead to a more predictable, lighter-touch regime for its next wave of AI.

Continue reading https://foxvector.com/stories/019f082e-fdaa-1309-732d-2f3bcfaf3451