In June 2026, the U.S. Commerce Department sent Anthropic a directive ordering it to block foreign nationals from accessing its most advanced AI models. Anthropic had no way to filter users by nationality in real time. So it did the only thing it could: it shut the models off for everyone on Earth. OpenAI is delaying its next-generation GPT-5.6 at the White House's request, releasing it first to a small set of government-preapproved customers. Meanwhile, across the country, numerous bills have been introduced this year telling schools when and how students may use AI at all. Somewhere in this pile of directives and delays, a student is trying to learn how the most capable tools of her generation actually work — and every institution in her life is conspiring to make sure she can't.
Here's what's happening, concretely. The federal government is building a system where frontier AI models go through a government benchmarking process before release, with developers voluntarily giving the government 30-day early access. The framework doesn't exist yet — OpenAI's executives said so explicitly. But the delays are already here. Anthropic's models got yanked. OpenAI's GPT-5.6 is being parceled out to "trusted partners" selected by the government. The executive order says this isn't a licensing regime. It is functioning exactly like one.
And who are these "trusted partners"? Defense contractors. Cybersecurity firms. Critical infrastructure operators. Not universities. Not students. Not the people who are supposed to be preparing for a world shaped by this technology.
Teach It but Don't Touch It
The absurdity runs deeper than federal policy. Several states have bills restricting AI use in classrooms or requiring oversight on student interactions. Many schools or districts don't allow students to use generative AI at all. And yet employers are increasingly requiring AI skills of job candidates.
Read those two facts back to back. Schools are banning the tools. Employers are requiring fluency with them. Nobody seems to have noticed these are the same students.
Universities are doing their own version of this dance. A review of major U.S. universities found that few have banned AI outright — but a significant portion leave AI policy entirely to individual instructors. Which means a student's access to the defining technology of the decade depends on whether their professor is curious or frightened. That's not a policy. That's a coin flip.
The gap between well-resourced and under-resourced institutions is widening. The schools with money buy enterprise licenses. The schools without money write bans. The students at the second kind of school pay the same tuition.
What makes this particularly galling is who does get access. When Anthropic kept advanced models restricted, the company shared them with vetted organizations through partnership programs. These partners were able to identify numerous high-severity vulnerabilities. The models are staggeringly powerful. And the people learning to build the next generation of software? Locked out.
OpenAI isn't happy about the GPT-5.6 delay either. The company wrote that this arrangement "keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them." They're right. But notice who didn't make OpenAI's list of people who matter: students.
This is what a capability ladder looks like when governments and corporations are the only ones allowed to climb it. Government agencies get the model. Financial institutions get the model. The student in a state school cybersecurity program gets a ban, a plagiarism warning, and a professor who hasn't updated their syllabus.
If you run a university and you're not screaming about this — about the fact that the government is choosing which companies get access to frontier AI while your students are told to put their phones away — then you've already decided your institution is a museum. A place where people study history. Not a place where they make it.