ThinkPatternGet the app
Perspective
TECHNOLOGY · AUG 14, 2026

The AI Enforcement State Points at Everything Except AI

The government can deploy AI to catch a misrouted shipping container but uses congressional hearings to investigate American companies running on Chinese AI models.

Customs and Border Protection's new AI system scans shipping data across more than 40 countries to catch Chinese exporters routing goods through third nations to dodge US tariffs. The system, called "detective border," flags transshipment networks in real time. The agency reports it hits 95 percent accuracy.

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

When the same government wanted to investigate whether American companies were running their operations on Chinese AI models, it did not deploy anything comparable. House committees sent letters to Airbnb and Cursor, the AI coding startup, demanding answers about their use of low-cost Chinese models [1]. The instrument was the same one Congress would use to investigate a pharmaceutical company's pricing. Behind the detective border sits a rapidly expanding enforcement infrastructure. In February, Palantir secured a $1 billion five-year blanket purchase agreement with the Department of Homeland Security, covering Customs and Border Protection, ICE, the TSA, FEMA, the Secret Service, and CISA. No competitive bidding was required [2]. ICE had already used Palantir's Falcon application for eight years to track subjects through driver's license scans and cell tower records; the agency has since expanded the relationship with a $30 million ImmigrationOS contract [3]. In June, Trump signed executive orders to integrate AI into the military and national security enterprise, directing the NSA to run classified benchmarking of frontier models and creating a Treasury cybersecurity clearinghouse [4]. CISA is piloting Anthropic's AI to scan federal software for vulnerabilities. The NSA uses the same system in classified settings [5]. DOGE and Virginia's state government under Governor Glenn Youngkin have deployed agentic AI to identify and cut regulations [6]. Palantir and NVIDIA launched a "Sovereign AI Operating System" marketed explicitly to defense ministries and central banks as a tool for running AI locally as a sovereign enforcement instrument [7]. Every one of these deployments targets the physical layer. Goods crossing borders. People moving through checkpoints. Software vulnerabilities in federal code. Regulations on the books. The state is building an AI enforcement apparatus that can see a misrouted shipping container moving through a transshipment network spanning forty countries. What it is not using AI to police is the flow of Chinese AI models into American enterprise software. That flow is substantial. A US-China Economic and Security Review Commission report found that 80 percent of US startups now use Chinese open-source base models [8].

This open ecosystem enables China to innovate close to the frontier despite significant compute constraints. — U.S.-China Economic and Security Review Commission

Chinese models process 12.96 trillion tokens per week against 3.03 trillion for US models, at roughly one-tenth the cost [9]. Pinterest uses DeepSeek R-1 for its recommendation engine. Airbnb runs Alibaba's Qwen on customer service. Pinterest reports the open-source techniques yield models 30 percent more accurate than off-the-shelf proprietary options [10]. And when Congress wants to know about it, it holds hearings [1]. The state that can deploy AI to catch a misrouted shipping container uses the same investigative instrument for Chinese AI models that it would use for any industry dispute. The administration is not unaware of the gap. It is divided over what to do about it. White House AI adviser David Sacks has called proposed restrictions on Chinese models "regulatory capture" designed to protect a "duopoly" of US closed labs.

the leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition. — David Sacks

A blanket ban faces a separate obstacle. The 1965 Supreme Court precedent Lamont v. Postmaster General, which protects the right to receive foreign materials, makes an outright prohibition legally infeasible. Open-weight models can be downloaded and hosted on private servers, putting them beyond any procurement-level control [11]. The recursion is the point. While the detective border AI scans shipping manifests for tariff evasion, American companies are using AI customs documentation tools to bypass those same tariffs. The Light Phone deployed such tools and cut its costs 20 percent [12]. The same technology sits on both sides of the same wall: one instance of it calculating the tariff, another calculating the way around it.


Sources
  1. 1. US and China Escalate AI Conflict Over Security and Exports
  2. 2. Palantir Secures $1 Billion Agreement with Homeland Security
  3. 3. ICE Used Palantir Technologies Inc. to Track Targets for Eight Years
  4. 4. Trump Orders Military Acceleration of Artificial Intelligence Integration
  5. 5. CISA Uses Anthropic AI to Scan Government Software
  6. 6. DOGE and State Governments Deploy AI to Cut Regulations
  7. 7. Palantir and NVIDIA Launch Sovereign AI Operating System
  8. 8. US Commission Warns China's Open-Source AI Threatens US Leadership
  9. 9. Chinese AI Models Outpace U.S. Rivals in Global Token Usage
  10. 10. U.S. Enterprises Adopt Chinese Open-Source AI Models
  11. 11. US Debates Legal Feasibility of Banning Chinese AI Models
  12. 12. US Companies Use AI to Offset Trump Administration Tariffs

Keep reading in the app

The full perspective, free in the app.

Download on the App StoreComing soonGoogle Play