The US Just Ripped a Chinese AI Out of Federal Systems on National Security Grounds. It Still Won't Regulate the Ones at Home.
// EDITOR'S NOTE
"WTF. I thought the big race was against China, not a collaboration with them?!"
We mean that literally, and the order proves it. This ban only exists because a Chinese AI model was already running inside American government IT — baked into software from American vendors, in the Treasury, State, and Justice departments. That is not two nations racing. That is one supply chain. The "race" is the ad copy; the procurement chain is the reality.
On Monday, the Department of Homeland Security made public an order that had been sitting since July 24: Chinese AI startup DeepSeek is banned from federal civilian agency IT systems. Agencies must remove DeepSeek-related technology — including third-party products and services built using the company's models — and are barred from procuring DeepSeek products or services going forward. The order applies to Treasury, State, and Justice, and excludes the Department of Defense.
The justification is blunt. DeepSeek's models "present multiple threat vectors," DHS wrote, citing potential access to sensitive data by Chinese authorities, the risk that U.S. government data could be transferred to China and used to build intelligence on American citizens and programs, and possible malware or other security vulnerabilities — including open-source research identifying multiple critical flaws in DeepSeek's iOS app.
Agencies can request exceptions, or a delay in compliance, by submitting a written justification to DHS within 60 days.
How a Chinese model got inside Treasury, State, and Justice
The scope of the ban is the part worth reading twice. DHS did not merely block DeepSeek's own app. It requires removal of technology "developed using the company's technology" — meaning American-built products sitting in federal systems were, at some point, running on a model built in Hangzhou. Federal civilian agencies bought that. Contractors handed it to them.
So after years of the AI race being framed as a contest between two blocs, the actual condition of American government computing was interdependence: Chinese open-source weights shipped inside U.S. vendor offerings, deployed into the departments that handle taxes, treaties, and prosecutions. The ban is not a firewall being erected around a rival. It is a dependency being audited after the fact.
Beijing's response was, of all things, a complaint about national security overreach. Foreign ministry spokesperson Mao Ning told Reuters that "China consistently opposes the over-generalization of national security as well as the politicization of science, technology and trade issues." Notably, the objection is framed around trade, not safety. Nobody in this exchange is arguing about whether the technology is good for people.
Sixty days to rip out a Chinese model. Two weeks ago, nobody could touch ours.
This is not a new posture, just a sharper one. The Defense Information Systems Agency moved to block DeepSeek in January 2025. Commerce Department bureaus banned it on government devices that March. Congress restricted it in intelligence community systems in December 2025. Governors in New York, Tennessee, and Virginia banned it on state devices. Monday's order adds the federal civilian government and a hard removal requirement.
Now set that beside what happened on October 5 in New York City, when the Council convened all 51 members and put the AI industry under oath. Jacob Coxon, a former Anthropic researcher, testified that on the current path it is "more likely than not" humanity loses control of these systems, and that the industry's approach is "extremely reckless." Daniel Kokotajlo, formerly of OpenAI, testified that the labs' ability to notice their own failures is poor and getting worse — that what passes for safety is "duct tape that will fall off later." Alex Turner, who left Google DeepMind over its Pentagon contract, put the odds of an AI takeover at roughly one in three, and said Google signed the deal while its own policy staff "never finished evaluating it."
Speaker Julie Menin's framing that day was that when the federal government fails to act, cities take the lead. This week, the federal government acted. It acted on the nationality of a chatbot, not on the failures the industry's own former employees described under oath fourteen days earlier.
And the "we can't do anything" argument died in the process. DHS can compel the removal of a foreign model from the Treasury, State, and Justice departments within 60 days and bar its procurement on data-security grounds. Whatever the constraints on regulating domestic labs are, they are not legal incapacity. They are preference.
Now, about the race
You have heard the race invoked in every county that has ever tried to slow a data center down. It arrives as a threat and a promise: we cannot pause, we cannot impose water limits, we cannot require an independent review of the numbers — because China will win. Local objections are reframed as national sabotage.
If that framing is true, then the national security stakes are exactly as serious as DHS just said they are. Which makes the no-brakes strategy indefensible. You do not win a security contest by stripping the guardrails off your own grid, your own aquifers, and your own farmland, and sending the bill to ratepayers. A country that treats a downloaded chatbot as a threat to the Republic is not somehow powerless to require a third-party audit before a 1-gigawatt facility gets a 20-year power contract.
And here is the collaboration, stated plainly. Both countries are building the same thing in the same way: the same accelerator chips, the same gigawatt campuses, the same drained aquifers, the same "move fast, break things" operating culture that the whistleblowers described under oath. The models get compared for a scoreboard; the concrete, the water, and the transmission lines are shared strategy. Two governments racing to the same cliff edge are not rivals. They are a convoy.
One more detail that undercuts the whole theatrical frame: DeepSeek's models are open source, and the lineup keeps shipping. You can remove them from federal procurement. You cannot remove the capability from the world — and the order's own scope admits the models already circulated beyond the federal civilian systems it covers, into state agencies, private companies, and vendor products. A ban that arrives after the integration is not a firewall. It is a confession about how much was already shared.
Say no to data centers. A foreign chatbot is now a matter of national security. A gigawatt facility on a strained grid is at least as much of one, and your county board is the only body that has to ask.
Sources & Credits
Primary: "U.S. bans DeepSeek from government agency systems amid AI rift with China," CNBC (Oct. 20, 2026).
Order: DHS notice on DeepSeek in federal civilian agency IT systems — issued July 24, 2026, posted for Federal Register publication Oct. 20, 2026.
Additional coverage: Bloomberg — "US Moves to Bar China's DeepSeek From Federal Agencies and Contracts" (Oct. 19, 2026); Politico — "DeepSeek barred from government agency IT systems"; BBC — "US bans China's DeepSeek from government in race for AI dominance"; GovInfoSecurity — "DHS Bans DeepSeek From Federal Civilian Agencies".
Prior restrictions: Reuters — Commerce Department bureaus ban DeepSeek on government devices (March 2025); Intelligence Authorization Act, FY2026 (P.L. 119-60), which restricted DeepSeek in intelligence community systems (Dec. 2025).
Industry testimony referenced: "New York City Just Wrote the AI Rulebook" — No More Data Centers coverage of the Oct. 5, 2026 City Council hearing.
// DID THIS RESONATE?
CLICK TO SHOW YOUR SUPPORT
