US Government Asks OpenAI and Anthropic to Limit Rollout of New AI Models Washington has quietly become the new product manager for frontier AI — and not everyone is thrilled about it.

OpenAI’s latest flagship, the GPT‑5.6 series, is being rolled out not to the public, but to “a small group of trusted partners,” and explicitly “at the request of the US government.” The company calls it a temporary, safety‑driven pause while it works with the administration on a repeatable review process for powerful models like Sol, Terra, and Luna. Washington’s logic: early, tightly controlled access lets officials probe risks from cyberattacks to military misuse before the tools go wide.

The government’s case: security first, markets later

From the government‑aligned view, OpenAI is simply complying with a national‑security playbook that began with Anthropic. After an export‑control directive, Anthropic was forced to disable access to its Mythos 5 and Fable 5 models just days after launch, in what officials framed as a necessary response to security threats. OpenAI’s “short‑term” arrangement is cast as a bridge while the administration builds a formal framework under a June 2 executive order on frontier AI review.

The opposition’s case: creeping control and a two‑tier internet

Critics see something darker: a de facto licensing regime emerging without legislation. OpenAI has handed over information on who gets early access, effectively turning its most advanced model into a government‑blessed walled garden. Anthropic tells a similar story: two weeks after being ordered to cut off all foreign users — including its own non‑US staff — from Mythos 5 and Fable 5, it now says the US has allowed it to reopen Mythos 5 only for “trusted” American organizations managing critical infrastructure and some Fortune 500 firms.

Where the government stresses “significant progress” and managed risk, opponents warn that ad‑hoc oversight, selective access and opaque security criteria are hardening into precedent — one restricted model at a time.