July 16, 2026

Frontier AI Now Has a DMV

Access to frontier AI is starting to look less like software onboarding and more like getting a driver’s license. That should make AI companies nervous.

For the past decade, the default growth motion was simple: let a customer sign up, run some basic checks, monitor usage, and ship. But that model is colliding with a new regulatory reality.

Governments are beginning to treat access to the most powerful AI systems as a privilege that can be granted, restricted, or revoked at any time.

What does restricting access to frontier AI actually mean?

Restricting access means regulators and governments are starting to regulate advanced models as permissioned resources. The government may require AI platforms to provide users’ proof of identity, purpose, and jurisdictional clearance before enabling use.

AI companies were not built for this. They were built for the opposite: frictionless, product-led growth. But now regulators are moving AI companies toward gated access, and the platforms that cannot prove who their users are will seriously stunt their ability to scale.

I learned the logic of licensing at 16. In Houston, “close by” is a relative term and nothing worth getting to is actually walking distance. Turning 16 wasn’t a milestone so much as a deadline. I was getting my license the day I turned 16, full stop, because there was no version of my life in that city that worked without one.

I wasn’t worried about the written test or the driving test. I was worried about parallel parking a pickup truck that felt roughly the size of the DMV itself, in front of a woman named Ethel who had clearly watched a thousand nervous sixteen-year-olds fail at this exact maneuver.

Twenty minutes, one written test, and one white-knuckled parallel-park attempt later, the camera caught the most important photo of my life so far: my first license.

It’s a good system when you think about it. Before anyone is let loose behind two tons of steel, some agency gets to decide whether they’re fit to operate it. And that decision is never really final. Get too many tickets, fail a renewal, or even forget your glasses, and the privilege gets pulled, no matter how badly you need to get somewhere.

That bargain makes sense for cars. The risk is visible and the state has spent a century getting reasonably good at deciding who belongs on the road. None of that is true for frontier AI.

How regulators are gating access to powerful AI models

In June 2026, regulators took their first real steps toward gating frontier AI access, and the industry hasn’t reckoned with what that means.

Under a U.S. export control directive, Anthropic temporarily suspended access to its Mythos 5 and Fable 5 models for non-U.S. users. Access was restored July 1, once the controls were lifted. Two weeks later, during the initial limited preview of GPT-5.6, OpenAI said access would be granted only to a small number of trusted organizations, with participants reviewed in coordination with the U.S. government on a customer-by-customer basis before broader availability. Both companies also gate access by use case under the EU AI Act.

Different mechanisms, same underlying function: regulators deciding, case by case and country by country, whether a given company gets to use a given AI system, and revoking that right whenever it sees fit.

That’s licensing through regulation, and the AI industry is not built for it.

There’s no consensus that gating a company’s access to its own customers makes anyone safer, and there’s little evidence regulators know how to make that call, model by model, country by country.

AI companies built their businesses assuming they could sign up a customer and ship. Now the onboarding decision may have to run through an agency that may or may not understand what it’s approving.

Why current onboarding lets dangerous actors slip through

Most AI platforms focus on catching suspicious behavior once someone is already inside: unusual usage patterns, rate limit abuse, jailbreak attempts.

It’s a guardrail most product-led growth platforms were built on, because that’s where their insights reached. And that’s where the biggest gap exists. Catching a jailbreak tells you what someone tried to do once they were already in. It says nothing about who they are.

That’s the gap regulators are reacting to, and the one platforms don’t have a defensible answer to.

How identity and KYC failures create national security risk

The scale of what’s coming is bigger than most platforms assume.

An exclusive Socure report on fraud rings targeting federal and state programs found that international groups originating from China, Russia, and elsewhere accounted for up to 12% of all applications for government services and loans, with one in four attempts hitting multiple agencies at once.

AI-driven tools are enabling foreign operators to spin up synthetic shell companies and identities that bypass standard KYC and business onboarding flows with ease and scale. A sophisticated fraud attack that once cost $20k to pull off now costs less than $50 *.

That’s the frightening pattern regulators are reacting to: foreign actors operating at industrial scale, against onboarding flows built to convert more customers without verifying who the users actually are.

What this means for enterprise growth and procurement speed

The impact goes beyond regulatory intervention. AI platforms are racing to embed deeper into enterprise workflows in highly regulated industries like financial services. Security and risk teams will also need to know who access is being granted to.

Whichever platform can operationalize these governance controls will be the ones enterprise buyers trust with their most important workflows.

The DMV doesn’t disappear because an industry hoped to outgrow it. It just decides who gets the keys, one case at a time, until someone builds the proof first: a verifiable, portable record of who’s actually on the other end.

In my next piece, I’ll show you what that looks like. But first, I’ll shed light on what scared the government into reaching for the keys: four documented cases from the past twelve months where fraud stopped being fraud and became a national security problem.