The U.S. Government Just Decided Who Gets Access to the Most Powerful AI Models
Two stories dominated AI news this week, and they’re really one story: the U.S. government is now actively deciding which Americans — and which companies — can use the most advanced AI tools. That’s a significant shift, and it happened fast.
OpenAI Delayed Its Own Model Launch Because Washington Asked
OpenAI released GPT-5.6 this week, but not the way it planned. At the direct request of the Trump administration, the company pulled back from a public launch and instead limited access to a small group of “trusted partners.” The administration cited national security concerns as the reason for the delay, according to The Verge.
Think of it like a pharmaceutical company finishing a new drug but agreeing to hold it back from pharmacies while regulators review who should get it first. OpenAI finished the product. The government said: not yet, not for everyone.
OpenAI didn’t exactly push back — but it did make its position clear. The company stated publicly that it doesn’t want access restrictions to become routine. Its specific concern is practical: cybersecurity professionals and international allies sometimes need cutting-edge AI tools to do their jobs. Locking those people out in the name of security could, paradoxically, create new security gaps.
Why this matters: If you were waiting for GPT-5.6, you may be waiting longer. More broadly, a new pattern is forming where government sign-off becomes part of every major AI release.
“Trump administration asked OpenAI to stagger release of GPT-5.6 over security concerns”
Anthropic’s Mythos 5 Is Now Available — But Only to 100 Selected Groups
Anthropic — the AI company behind the Claude family of models — has a new flagship model called Mythos 5. Weeks ago, an export control directive forced Anthropic to take it off the market entirely. This week, after negotiations with the Trump administration, access was partially restored. More than 100 U.S. companies and government agencies can now use it, as TechCrunch reports.
Export controls are laws that restrict who can access certain technologies, usually to prevent them from reaching foreign adversaries. They’ve historically applied to weapons and semiconductors. Applying them to AI software is newer territory, and this situation shows how quickly that can play out in practice.
For the 100-plus organizations now cleared to use Mythos 5, this is good news. For everyone else, it means one of the most capable AI systems available right now is simply not an option. If your company wasn’t on the approved list, you don’t get in. There’s no public signup, no waitlist, no workaround.
Why this matters: The U.S. government is now functioning as a gatekeeper for which businesses can access frontier AI — meaning the most advanced AI available at a given time. That’s a new kind of market power, and it’s worth watching closely.
“Mythos 5 model cleared for select group of companies and agencies”
OpenAI Draws a Line It Doesn’t Want Crossed
OpenAI isn’t just complying with the GPT-5.6 restrictions silently. As TechCrunch covered, the company went out of its way to say this kind of intervention shouldn’t become normal. That’s a careful position — cooperative enough to avoid a fight with Washington, but clear enough to signal where OpenAI thinks the line should be.
The underlying tension is real. AI companies depend on widespread adoption to generate revenue and stay competitive. Restricting who can use a model doesn’t just slow the technology’s spread — it directly affects the business. OpenAI is essentially arguing that a less open AI ecosystem makes everyone less safe, not more.
For everyday users, the immediate effect is simple: you probably won’t have access to GPT-5.6 anytime soon unless you’re using it through a trusted business partner or institution. The longer-term effect is harder to predict. If this becomes the norm, future AI releases may routinely go through government review before reaching the public.
Why this matters: OpenAI is threading a needle between cooperation and objection. How that balance shifts in the next few months will shape how AI products reach consumers for years.
Also Happening in AI
South Korea announced plans to train all 500,000 of its active-duty military personnel to operate drones, per Ars Technica — a striking example of how AI-adjacent technology is being embedded into national defense at scale. On the business side, OpenAI made a significant leadership hire: Prabhjeet Singh, who previously ran Uber’s operations across India and South Asia, will now lead OpenAI’s India strategy, its largest market outside the United States, TechCrunch reports. And for those tracking OpenAI’s full product update: GPT-5.6 actually launched as three models — a flagship version called Sol and two smaller variants — even as the government access drama played out in parallel, according to The Verge.
What to Watch
The real story isn’t any single model or delay — it’s that the U.S. government has established a working pattern of intervening in AI releases before they go public. Watch for whether other AI labs face similar requests, and whether Congress moves to formalize this process into actual law. If it does, what’s now informal negotiation between tech companies and the White House could become a regulated approval process — one that permanently changes how quickly new AI reaches ordinary people.