Item: OpenAI
Author: Karl Strauchman

The Trump administration on Wednesday asked OpenAI to restrict GPT-5.6 to a small set of government-approved partners before any wider rollout, the first time the U.S. federal government has preemptively gated a frontier model launch. Axios first reported the request, which came from the White House's Office of the National Cyber Director and the Office of Science and Technology Policy and was triggered, per a source familiar with the matter, by what the administration described as the model's "Mythos-like" capability.

That label is doing real work. Earlier in June, Trump signed an AI security executive order standing up a voluntary pre-release testing protocol across several agencies. Voluntary is the operative word: GPT-5.6 is the first model where the protocol has been invoked in anger, and Commerce Secretary Howard Lutnick reportedly wanted relevant parts of the government to have tested and approved the model before release. Sam Altman discussed it with Lutnick the same day OpenAI was on stage with Broadcom unveiling its first custom chip.

In a staff memo first reported by The Information, Altman told employees: "We've made clear to the U.S. government that this is not our preferred long term model, and will work with them and others in industry to achieve a more sustainable approach for future releases." The phrasing is careful. OpenAI is complying, and signaling that it doesn't want compliance to become the default posture.

The chip announcement, made Tuesday, is the other half of the story. Jalapeño, a custom LLM inference ASIC OpenAI is calling an "Intelligence Processor," went from initial design to tape-out in nine months, which OpenAI says is possibly the fastest cycle in high-performance semiconductors. Engineering samples are already running production workloads, including GPT-5.3-Codex-Spark, at target frequency and power, with performance-per-watt OpenAI claims is "substantially better than current state-of-the-art." Celestica is handling board and rack integration; Broadcom's Tomahawk silicon stitches the systems together. OpenAI says its own models assisted parts of the design work.

Greg Brockman, OpenAI's president, framed that recursion plainly: "The degree to which our models have been able to accelerate it was very surprising to us."

Broadcom CEO Hock Tan described "small prototype development" in late 2026, then "really ramp up in '27 and really going full tilt in first half '28," with gigawatt-scale deployments alongside Microsoft and other partners. "This is just the beginning of a multi-generation roadmap," he said in the joint announcement. Broadcom shares are up roughly 10% year-to-date.

The juxtaposition is the news. On Tuesday, OpenAI announced the silicon supply chain underwriting frontier-model scale through 2028. On Wednesday, Washington pulled the launch lever on the model that supply chain exists to serve. Compute capacity is going private and proprietary; the right to deploy it's becoming, for the first time, federally permissioned.

OpenAI ships Jalapeño as Washington gates GPT-5.6 behind federal approval

Sources