There you have it! A federal appeals court just told the Pentagon it can keep the best AI company…
September 26, 2026 · 0 likes · 0 comments
China Threat Defense Cybersecurity AI
There you have it! A federal appeals court just told the Pentagon it can keep the best AI company in the world off our own military's systems.
The D.C. Circuit ruled 2-1 that the Department of Defense was right to blacklist Anthropic as a "supply-chain risk."
Read the reason, because it will make your blood boil.
Anthropic didn't leak anything. It didn't get hacked. It didn't sell to Beijing.
Its crime was writing two lines into its usage policy:
Don't use Claude to build fully autonomous weapons.
Don't use Claude for mass domestic surveillance of Americans.
That's it. That's the "national-security risk."
Those aren't obstruction. Those are the guardrails any sane engineer would want on the most powerful technology ever built. Every serious company draws lines like these. They are standard. They are reasonable. No honest person reads "no autonomous kill-decisions, no spying on citizens" and calls it a threat.
And for that, Emil Michael, the real national security threat, declared one of the two labs actually winning the AI race a "supply-chain risk" — and just got a court to bless it.
Let me be honest about what this is.
Anthropic came to the table. They wanted to work with the Pentagon. They built Claude into government systems. They asked for two commonsense limits and offered everything else.
We answered by throwing them off the island.
Meanwhile China's labs don't write usage policies. They write targeting packages. Their "guardrails" are whatever the Party says on a Tuesday. That is who we are handing the advantage to when we punish an American company for having ethics.
Judge Henderson dissented. She was right. The majority found "ample support" for a risk that anyone outside a windowless procurement office can see doesn't exist.
Here's the part nobody in that building wants to say out loud: this was never about security. A San Francisco court already looked at a parallel designation against the same company and ruled it unlawful. Same facts. Opposite verdict. One of these rooms is wrong, and it isn't the one that read the policy.
We are in the fight of the century for AI supremacy, and we just spent a court case kicking one of our own best players out of the arena — over a sentence that says don't build a killer robot without a human in the loop.
Shameful.
Fix the procurement process. Fix the officials who confuse safety terms with sabotage.
A normal, healthy person would just have stopped using it. Instead, Emil's fragile ego was hurt and he retaliated and banned them.
Now Genai.mil runs on Gemini (useless) and GPT 5.4.. and OpenAI just released a DoW and IC model with customized guardrails.. running GPT 5.4. Useless.
Full story: https://lnkd.in/eTjSmyej
Would you rather our military run on the best American AI with guardrails — or on whatever has none?
The D.C. Circuit ruled 2-1 that the Department of Defense was right to blacklist Anthropic as a "supply-chain risk."
Read the reason, because it will make your blood boil.
Anthropic didn't leak anything. It didn't get hacked. It didn't sell to Beijing.
Its crime was writing two lines into its usage policy:
Don't use Claude to build fully autonomous weapons.
Don't use Claude for mass domestic surveillance of Americans.
That's it. That's the "national-security risk."
Those aren't obstruction. Those are the guardrails any sane engineer would want on the most powerful technology ever built. Every serious company draws lines like these. They are standard. They are reasonable. No honest person reads "no autonomous kill-decisions, no spying on citizens" and calls it a threat.
And for that, Emil Michael, the real national security threat, declared one of the two labs actually winning the AI race a "supply-chain risk" — and just got a court to bless it.
Let me be honest about what this is.
Anthropic came to the table. They wanted to work with the Pentagon. They built Claude into government systems. They asked for two commonsense limits and offered everything else.
We answered by throwing them off the island.
Meanwhile China's labs don't write usage policies. They write targeting packages. Their "guardrails" are whatever the Party says on a Tuesday. That is who we are handing the advantage to when we punish an American company for having ethics.
Judge Henderson dissented. She was right. The majority found "ample support" for a risk that anyone outside a windowless procurement office can see doesn't exist.
Here's the part nobody in that building wants to say out loud: this was never about security. A San Francisco court already looked at a parallel designation against the same company and ruled it unlawful. Same facts. Opposite verdict. One of these rooms is wrong, and it isn't the one that read the policy.
We are in the fight of the century for AI supremacy, and we just spent a court case kicking one of our own best players out of the arena — over a sentence that says don't build a killer robot without a human in the loop.
Shameful.
Fix the procurement process. Fix the officials who confuse safety terms with sabotage.
A normal, healthy person would just have stopped using it. Instead, Emil's fragile ego was hurt and he retaliated and banned them.
Now Genai.mil runs on Gemini (useless) and GPT 5.4.. and OpenAI just released a DoW and IC model with customized guardrails.. running GPT 5.4. Useless.
Full story: https://lnkd.in/eTjSmyej
Would you rather our military run on the best American AI with guardrails — or on whatever has none?