The Administration Now Describes Tariff Evasion and Domestic Dissent in the Same Language

When the White House released a report this week on Chinese tariff evasion, it chose a phrase that would have been at home in a counterintelligence briefing ^[1].

Preferential access to the American market is not a license to launder somebody else's exports. — Peter Navarro

The report was about shipping containers and falsified customs forms. But the grammar — the idea of a clandestine network spanning dozens of jurisdictions, requiring mapping and dismantling — is the same grammar the administration has been using to describe a very different set of targets. In the past year, DHS has run an operation called "Puppet Master" that infiltrated labor union Signal chats, recorded meetings at churches and libraries, and used facial recognition to build dossiers on progressive activists ^[2]. The Treasury Department has opened investigations into American activists for "coordination" with the Cuban regime, treating domestic political organizing as a foreign-influenced network ^[3]. USCIS now screens legal immigration applicants' social media for "anti-American" views, treating such activity as an "overwhelmingly negative factor" that can justify denial ^[4]. The convergence is linguistic before it is technical. The administration is describing two seemingly separate problems — Chinese trade evasion and domestic political dissent — in the same sentence structure. And that sentence structure points to shared infrastructure underneath. The most concrete layer is procurement. In February, Palantir secured a five-year, $1 billion blanket purchase agreement with the Department of Homeland Security ^[5]. The agreement lets six agencies — Customs and Border Protection, Immigration and Customs Enforcement, the TSA, FEMA, the Secret Service, and the Cybersecurity and Infrastructure Security Agency — buy Palantir software without competitive bidding. Under one contract, CBP can purchase the same vendor's tools for its trade-enforcement mission that ICE buys for deportation operations and domestic surveillance. The BPA does not dictate which Palantir applications each agency runs, but it removes the procurement friction that would otherwise keep those missions in separate technological lanes. One floor down is the technical authorization layer. In February, the Defense Information Systems Agency authorized Palantir's full software stack — Gotham, Foundry, Apollo, and AIP — for Top Secret edge deployments under a process called "Authorize Once, Use Many" ^[6]. The rule means any government customer can deploy the same Palantir suite across any environment, from enterprise data centers to tactical field operations, on hardware of their choosing. A DISA spokesperson made the scope explicit.

can be deployed across any environment, from enterprise data centers to the tactical edge, on hardware of the customer's choosing. — Palantir

A month later, Palantir and NVIDIA packaged the same stack as a "Sovereign AI Operating System" — a turnkey product combining NVIDIA Blackwell hardware with Palantir's full software suite, marketed to governments as infrastructure for running intelligence and surveillance operations on local servers without sending data to external cloud providers ^[7]. The product name is the pitch: a government's own AI, running on its own machines, answerable to its own rules. Beneath the technical layer is the legal one. A single directive — National Security Presidential Memorandum 7 — ordered a "whole-of-government approach to combat left-wing political violence" and expanded the definition of domestic terrorism ^[2]. That memo is the authority cited for Operation Puppet Master's infiltration of labor unions and progressive groups. It is also the authority cited for the administration's push to consolidate all foreign and domestic intelligence targets into a single centralized database maintained by the Office of the Director of National Intelligence — what critics call a master spy list ^[8]. One presidential memorandum supplies the legal spine for both the domestic surveillance of political organizations and the unification of intelligence targeting across agencies. The administration has not only built the stack; it has dismantled the mechanisms that would constrain it. DHS shuttered its Office for Civil Rights and Civil Liberties ^[9]. The Justice Department created an Artificial Intelligence Litigation Task Force to preempt state-level AI safety laws, even as California and Texas enacted their own ^[10]. Attorney General Pam Bondi pressured Apple to remove ICEBlock — a crowdsourced app for documenting ICE raids — from the App Store ^[9]. When Anthropic raised ethical concerns about mass surveillance and the use of its Claude model in classified military networks, the Pentagon retaliated by labeling the firm a supply chain risk, terminating a $200 million contract, and barring other contractors from working with it; Anthropic is now suing for illegal retaliation ^[11]. Meanwhile, the administration's own AI security framework — signed by executive order in June — is explicitly voluntary, prohibiting any mandatory licensing or permitting requirements for government AI deployments ^[12]. The asymmetry is the point: AI used by the government faces no binding constraint; AI firms that resist integration face consequences. The architecture is real and expanding, but it is not seamless. Federal courts have issued multiple injunctions: Judge Talwani blocked IRS-ICE data sharing in February after the IRS had already turned over 47,000 taxpayer records to immigration authorities ^[13]. The CIA and FBI are actively resisting the master spy list, with senior counterintelligence officials warning that consolidating compartmentalized data into a single database creates "catastrophic risks" and that one leak could compromise covert operations ^[8]. And ICE's own arrest data undercuts the justification the administration has offered for its deportation campaign: in Los Angeles, 39% of detainees had no criminal charges or convictions; in Minnesota's Operation Metro Surge, the figure exceeded 60% ^[14].

Many of the individuals that are counted as ‘non-criminals’ are actually terrorists, human rights abusers, gangsters and more; they just don’t have a rap sheet in the U.S. — U.S. Department of Homeland Security

ICE denies that its surveillance apparatus targets political speech. The agency has stated that its social media monitoring is designed to investigate threats against its personnel, not to monitor dissent.

Any allegation DHS and ICE are attempting to ‘squash’ free speech is categorically FALSE. — United States Immigration and Customs Enforcement

But the 39-to-60% figure sits in tension with that denial in a way no assertion can resolve. When a surveillance architecture processes enough people that a majority of those swept up in a major operation have no criminal record, the architecture is not doing targeted threat assessment. It is doing something broader — and the vocabulary of shadow networks and whole-of-government approaches is broad enough to accommodate it.