🤚 The Open-Palm Clearance

In a development that reads like a spy thriller ghostwritten by a compliance attorney, the White House has personally approved Anthropic to supply its Claude Mythos model to the National Security Agency — overriding the Pentagon’s own classification of the company as a supply chain risk.

White House Chief of Staff Susie Wiles signed off on the arrangement, according to the New York Times, because America’s intelligence agencies have a problem that no amount of classified memos can solve: they don’t have enough GPUs.

The deal includes a clause barring the model from processing Americans’ data — which is reassuring in the same way that a “Wet Floor” sign is reassuring in the middle of a hurricane. Key facts:

• Mythos can run on older Nvidia chips, making it the only viable option for classified networks starved of Grace Blackwell processors
• The Pentagon had previously flagged Anthropic because it refused to release its technology for “any lawful use”
• The final contract quietly drops the controversial “lawful use” language entirely
• The White House has also approved $9 billion in new AI chip procurement — pending Congressional approval, which is Washington’s way of saying “eventually”

👐 The Two-Handed Security Briefing

Let’s appreciate the exquisite irony here. Anthropic — the company that built its entire brand on AI safety, responsible deployment, and saying no to things — just became the NSA’s AI vendor of choice precisely because it said no to things.

The Pentagon called Anthropic a supply chain risk. The White House called the Pentagon’s bluff. And the reason Anthropic won this contract isn’t because Mythos is the smartest model (though it did find 10,000 vulnerabilities in open-source code last month). It’s because Mythos runs on older hardware. The intelligence community’s classified networks are running chips that commercial AI labs abandoned two generations ago. OpenAI’s models need the latest silicon. Google’s models need the latest silicon. Anthropic’s model needs a chip from 2023 and a firm handshake.

This is the AI equivalent of winning a government contract because your software runs on Windows XP. It’s not glamorous. It’s strategic.

The “lawful use” dispute is particularly delicious. The Pentagon wanted Anthropic to agree that its models could be used for “any lawful use” — a phrase so broad it could cover everything from intelligence analysis to generating PowerPoint slides about intelligence analysis. Anthropic refused. The White House eventually negotiated around the language rather than forcing compliance, which tells you everything you need to know about who has leverage in 2026: it’s the company with the model that fits on your existing rack.

🌿 The Gentle Awakening

There is something quietly profound about the fact that the most powerful intelligence agency on Earth is constrained not by policy, not by law, not by ethics — but by chip availability.

The NSA can intercept communications across continents. It can break encryption that would take consumer hardware millennia to crack. But it cannot run the latest AI models because it doesn’t have enough of the right GPUs. The agency that pioneered signals intelligence is now, functionally, in the same position as a college student trying to run Stable Diffusion on a laptop from Best Buy.

And so the White House writes a $9 billion check — or rather, asks Congress to write a $9 billion check — to fix a supply chain problem that exists because every organization on Earth is trying to buy the same chips from the same company at the same time. Nvidia isn’t just a chipmaker anymore. It’s a bottleneck with a stock price.

👑 The Gold-Leaf Intelligence Assessment

This deal restructures the relationship between AI companies and the national security apparatus in ways that will echo for years. Anthropic has established a precedent: you can be the safety-first company and the defense contractor, as long as you negotiate your own terms. The “lawful use” language is gone. The data restrictions are in place. The model runs on legacy hardware. It’s the most Anthropic outcome imaginable — principled, pragmatic, and slightly smug about it.

For the broader industry, the message is clear: the government doesn’t just want AI. It needs AI. And it needs it badly enough to override its own Pentagon, negotiate with a company it classified as a risk, and ask Congress for $9 billion in emergency chip money. The leverage has shifted. The AI companies aren’t petitioning the government for contracts anymore. The government is petitioning the AI companies for access.

Meanwhile, OpenAI and Google are presumably watching this unfold while their models sit on the civilian side of the air gap, waiting for hardware that doesn’t exist yet in the quantities required. Mythos didn’t win on benchmarks. It won on compatibility. In the classified world, the best model isn’t the smartest one — it’s the one that actually turns on.

“The safety lab said no to everyone, the Pentagon said no to the safety lab, and the White House said yes to both of them simultaneously. This is not a contradiction — it’s Washington.” — The Slap of Wisdom National Security Desk, filing this report from a SCIF that runs on vibes and a 2023 Nvidia A100