OpenAI just told the Trump administration it's going to let its newest model out — but only after the government gives each customer a personal stamp of approval. The company's new GPT-5.6 Sol model, which it says is better at finding and fixing security vulnerabilities than launching cyberattacks, will be restricted to roughly 20 approved customers during a preview period. No names have been disclosed.
Sam Altman reportedly told staff that the government is "approving access customer by customer" during this preview window, with a broader public release planned for "a couple of weeks later." The agencies driving this request: the Office of National Cyber Director and the Office of Science and Technology Policy. Altman himself spoke with Commerce Secretary Howard Lutnick about the release Wednesday as part of a series of negotiations between AI executives and Trump officials over recent weeks.
It's an unprecedented situation. A U.S. government isn't supposed to be picking which companies get access to a private company's product, one customer at a time. And OpenAI knows it.
"Should Not Become the Norm"
Here's where it gets interesting. OpenAI didn't just comply quietly. The company issued a statement that was unusually direct for a Silicon Valley firm under government pressure:
"We don't believe this kind of government access process should become the long-term default," OpenAI said. The company framed the vetting as a temporary step on the "path to broader availability in the coming weeks."
OpenAI also drew a line about its own model, saying Sol doesn't cross the company's internal risk threshold. But it acknowledged there could be unforeseen risks — especially if Sol gets paired with other tools. "That uncertainty, along with the model's broader step change in capabilities, is why we are pairing the model's increased capabilities with stronger safeguards and a phased release," the company said.
Let's be honest here: that statement reads like OpenAI is trying to have it both ways. Comply with the government while simultaneously signaling that this isn't how things should work. It's a tightrope walk, and the company clearly doesn't want this becoming routine. Because if it does, every AI release from here on out needs a White House green light before it hits the market. That changes everything about how these companies operate.
The Executive Order Framework
This didn't come out of nowhere. Earlier this June, Trump signed an executive order establishing a framework for the federal government to vet national security risks of advanced AI systems for up to 30 days before public release. The order described participation as "voluntary" — but voluntary in practice means something very different when the White House is actively asking companies to slow down.
The framework hasn't been fully developed yet. That's a detail worth holding onto: the rules of this game are being written as it's being played. And while Trump originally positioned his administration as "hands-off" on AI, the pivot toward federal oversight has been steady and deliberate over recent months.
The timing matters. This executive order came just weeks after Anthropic's Mythos model sparked genuine alarm about AI-powered cyber capabilities. The government was looking for a mechanism to act, and the executive order provided one — even if its voluntary framing makes enforcement murky at best.
Why the Government Got Nervous
You can't understand what's happening with OpenAI without understanding what happened with Anthropic. The company unveiled two models — Fable 5 and Mythos 5 — and immediately ran into trouble. Mythos, in particular, raised eyebrows for its ability to find software flaws at speeds no human analyst could match. Malicious hackers could theoretically weaponize that capability against critical computer networks worldwide.
Anthropic took both models offline days after unveiling them to comply with a Trump directive blocking use by foreign nationals. David Sacks, who co-leads Trump's council of technology and science advisers, was blunt about what happened: "Dario came to Washington a few months ago, back in April, and basically said that he had created a cyber weapon called Mythos. And he spiked the cortisol level, got everyone really worried."
Sacks added there was "some truth to it" regarding Mythos's advanced cyber capabilities. The government then lifted restrictions on Mythos 5 on Friday, enabling its redeployment to "a small group of cyber defenders and infrastructure providers." But Fable 5 — which Anthropic had pitched as a safer version — remained unavailable for two full weeks, even after Mythos restrictions were lifted. That inconsistency has drawn serious criticism.
Critics Push Back Hard
The backlash from across the political spectrum has been swift and unusually unified.
Rep. Lori Trahan (D-MA), co-author of a bipartisan AI regulation bill, said she's concerned "the Trump administration is deciding company by company who gets access to the newest AI model. No law. No process. No oversight. Just appointees in Washington deciding who's in and who's out."
Stanford cybersecurity expert Alex Stamos went further. After reviewing an analysis of Fable by Anthropic's primary cloud backer Amazon, Stamos said he found "no risks that aren't present with other publicly available AI models, including those made in China." His verdict: "Pretty much nobody in the cybersecurity industry believes there's any factual basis for this action."
Stamos, who previously served as chief security officer at Meta, didn't stop there. "If the administration is honest about wanting the United States to beat China in this race, then this is about the dumbest thing they could possibly do," he said.
Brad Carson, who runs a bipartisan pro-AI safety super PAC, described the current federal oversight as an "ad hoc, personalized, opaque, possibly lawless approach" lacking transparency and basic fairness. The language is unusually strong from someone in the AI safety community — which traditionally leans toward caution rather than confrontation.
The IPO Wildcard
There's another layer to this that most coverage has overlooked. Both OpenAI and Anthropic have been exploring moves toward going public on Wall Street, following SpaceX's record-setting IPO on June 12. Government oversight of model releases adds a serious complication to those plans.
Trump has floated the possibility of the U.S. government owning a stake in leading AI companies — describing a concept where "pieces could be given to the American public, where the American public essentially becomes a partner with the companies." That's not just oversight anymore. That's partial ownership. And it changes the entire calculus for companies considering an IPO.
Anthropic's relationship with this administration has been particularly fraught. The Pentagon designated Anthropic as a national security risk after the company raised ethical concerns about AI usage in warfare. Trump ordered federal agencies to stop using Claude. Anthropic responded with a lawsuit that's still working its way through federal courts.
OpenAI, by contrast, has maintained a more cooperative posture — Altman speaking directly with Commerce Secretary Lutnick, working closely with government agencies on release protocols. But cooperation comes at a cost. Every day the company spends navigating this approval process is a day it's not shipping product, building features, or preparing for public markets.
The question hanging over all of this: is this a one-time exception driven by genuine security concerns, or the template for how AI development will work in America going forward? OpenAI's statement suggests the company hopes it's the former. The government's actions suggest otherwise.