The Pentagon's AI targeting system Maven went from a $480M prototype to a $13B contract ceiling. Palantir built it, 20,000 analysts use it daily, and the same playbook is now moving to the border and the police
An analysis of US and EU AI surveillance procurement shows the Pentagon's Maven system (built by Palantir) growing from a $480M prototype to a $13B contract ceiling, with similar data-fusion and autonomous detection systems now expanding to border control, ICE, and policing.
The Pentagon paid $480 million for a prototype. The contract ceiling is now $13 billion. This isn't a tank or a missile. It's software that helps decide what counts as a target. And here's the uncomfortable part: it can't realistically be switched off anymore. I went through a report on US and EU AI surveillance procurement for 2024–2026, built on SAM.gov and EU TED tender data. It's more alarming than it looks. First. Governments stopped buying hardware. They're buying data fusion platforms. Maven, built by Palantir, is used daily by 20,000 military analysts, with up to 20 billion tokens a day running through it. Disconnecting from a system like that becomes practically impossible. Second. A border without people. Anduril towers for CBP: a $2 billion contract, $818 million spent so far. The towers detect and classify a person on their own from up to 2.8 km (about 1.7 miles). The operator only gets shown "anomalies." Third. Anonymity is disappearing. ICE paid Clearview AI $9.2 million for access to a database of 50+ billion images. A face in a crowd gets tied to social media profiles. Fourth. Europe is no better. Frontex is putting €184.3 million into heavy surveillance drones over the sea, with the data feeding into EUROSUR. A boat gets intercepted by another country's coast guard, and formally there's no pushback. Fifth, and the biggest paradox. The EU AI Act bans social scoring and mass real-time facial recognition in public spaces (with carve-outs for law enforcement), but classifies systems like EES and ETIAS only as "high risk." Compliance costs millions, so only the big players can afford it. A law meant to protect citizens ends up building a moat around monopolies. Who wins? Palantir and Anduril. Gross margins of 40–45% versus 8–10% for the traditional defense giants. They don't sell a product, they rent out an algorithm. The government becomes a permanent tenant. The direction is clear: from reacting to crime to predicting it. Tech battle-tested in war moves to the border. Then to policing. It gets tested first on people with no voice. Then they come for everyone else.
Pentagon investigators found that overreliance on Palantir's Maven Smart System AI contributed to a U.S. strike that killed 123 Iranian children, citing failures in intelligence data and targeting processes.
The Pentagon is shifting from Anthropic's Claude to a multi-vendor AI strategy after a contract dispute over lawful operational use, now evaluating models from OpenAI and Google Gemini for military deployments.
The Trump administration has awarded Palantir an initial $3.9 million contract to develop surveillance tools for tracking federal employees' return-to-office compliance at the USDA, with similar programs planned at the VA and SSA, raising privacy and worker rights concerns.
Major European banks and asset managers have significantly increased their investments in Palantir despite the company's controversial ties to human rights violations and military operations. The investment surge is largely driven by Palantir's rising stock price and its inclusion in major market indices.