Anthropic’s most advanced AI, the Mythos model, was supposed to be locked down tight. Export controls. National security. All that noise. But this week, the Trump administration quietly pried open the door — just a crack — for a handful of companies and government agencies to get their hands on it.
The move is a reversal of sorts. Last year, the Commerce Department forced Anthropic to disable Mythos 5 and its predecessor, Fable 5, citing unspecified threats to national security. Now, under the radar, the White House has authorized limited release. The terms? Shady. The criteria? Vague. The message? Clear: if you’re connected, you’re in.
The Mythos Mirage
Let’s be real here. Mythos isn’t some open-source toy. This is the model that could supposedly write code, analyze intelligence, and draft policy memos — all while sounding eerily human. Anthropic marketed it as the safest frontier model ever. Then the government decided it was too dangerous for export. But somehow, it’s safe enough for select domestic hands? Make it make sense.
Reports from Bloomberg and Reuters confirm that the approved entities include a mix of defense contractors, energy giants, and three federal agencies. No names, of course. Transparency is so last decade. What we do know: these are organizations that have lobbied hard for access, and have the right connections in the West Wing.
“They picked winners and losers in the AI arms race,” said a former Commerce official who requested anonymity. “This isn’t about security — it’s about favoritism.”
What’s the Catch?
Anthropic isn’t just giving away the keys. Sources say each recipient must sign a binding agreement that includes on-site audits, usage caps, and mandatory quarterly reports. But who enforces that? The company? The government? Neither has a stellar track record with oversight. Remember how well Facebook’s data agreements worked out?
The Mythos model is powerful. It can generate synthetic data, simulate economic scenarios, and even write malware detection scripts. That’s the kind of dual-use tech that keeps security experts up at night. But the administration’s calculus seems to be: better to have it under our watch than let it leak to Beijing.
No one’s arguing with that logic entirely. But the selective rollout smacks of a two-tiered AI economy — one for the insiders, another for the rest of us. Startups and smaller firms get locked out while old-guard contractors feast.
A Pattern of Secrecy
This isn’t the first time the Trump admin has played favorites with cutting-edge tech. Last year, they granted early access to quantum computing prototypes to a defense consortium. Before that, it was a special waiver for chipmaker Intel. The pattern is consistent: a closed-door meeting, a nod from an undersecretary, and suddenly the rules don’t apply.
Critics argue this undermines the very export controls meant to protect national security. If the government selectively releases the most advanced AI, what’s to stop a rogue employee at an approved firm from walking out with a thumb drive? Nothing. Absolutely nothing.
The irony is thick: the same administration that railed against “big tech monopolies” is now handing them exclusive access to the world’s most potent AI. Anthropic becomes the gatekeeper, the government becomes the bouncer, and everyone else waits outside.
The Fallout
So what happens now? Expect a flurry of lawsuits from companies that got left out. Expect congressional hearings where no one says anything new. And expect more leaks — because when you lock something this powerful in a room with a few dozen people, secrets don’t stay put.
For Anthropic, this is a lifeline. The company has been bleeding cash trying to comply with export rules. Now they can charge premium rates for “government-grade” access. It’s a win for their bottom line. For the tech industry, it’s a warning: if you want access, better make friends in high places.
None of this is illegal. It’s just… unseemly. In a time when AI is reshaping everything from warfare to medicine, decisions this consequential should be debated in public, not handed down in backroom deals.
“The genie is out of the bottle,” said an AI ethics researcher at MIT. “But now we’re deciding who gets to rub the lamp. That’s not policy — that’s patronage.”
What’s Next?
The administration says the program is a pilot. It will “monitor outcomes” and “adjust as needed.” Translation: if no disaster happens in three months, they’ll expand the list. By year’s end, expect a dozen more companies to join the club. By 2028, the whole concept of export control may be moot.
In the meantime, if you’re not in the club, you’re playing catch-up. The AI race just got a new rule: it’s not who builds the best model — it’s who gets to use it.
