Palantir Promised in Writing It Would Not Build for ICE. Then It Did.

In October 2020, Palantir put a promise on paper. Asked by Amnesty International how it handled the risk of its software being used against migrants, the company answered that it had 'purposefully declined to take on contracts with ERO and CBP' - the enforcement arms of immigration policing - because of the 'risks of disproportionate immigration enforcement' and 'potential serious human rights violations against migrants, refugees, and asylum seekers.'¹ It was an unusually clean line for a company that sells to spies and soldiers: here is a door we have chosen not to walk through. Five years later, Palantir walked through it.

The story told in the headlines is that Palantir got caught - a quiet surveillance contractor exposed building deportation machinery for ICE. That framing misses what actually makes the case damning. Palantir wasn't caught crossing a line nobody knew about. It crossed a line it had drawn itself, in writing, for the record. The controversy isn't a secret revealed. It's a reversal documented.

The thing it built was the thing it said it wouldn't

In April 2025, under Trump's second term, ICE awarded Palantir a $30 million contract to build ImmigrationOS - a platform designed to give the agency 'near real-time visibility' on self-deportations, to track visa overstays, and to streamline deportation logistics. The contract justification named Palantir as the sole source capable of delivering a working prototype by September 25, 2025.² Read the 2020 letter and the 2025 contract side by side and there is no need for interpretation. The category Palantir said it had declined - enforcement against migrants - is the category it now serves. The human-rights risk it named as the reason for declining is the risk a deportation-logistics engine is built to act on. The reversal isn't subtext. It's the spec sheet.

And the $30 million headline number badly undersells the relationship. ImmigrationOS is one line item. Palantir's investigative case-management work for ICE goes back to an initial contract in 2014 worth over $41 million, with the current case-management system - a separate $139.3 million award in 2022 - slated for sole-source renewal because ICE could not find a comparable system from any other vendor.³ That last clause is the whole mechanism in a sentence. Once an agency's investigations run on your software, no rival can be 'comparable' - the data, the workflows, and the institutional muscle memory are yours. The lock-in that makes Palantir indispensable to ICE is the same lock-in that makes 'we'll just decline future contracts' a far more expensive promise to keep than it was to make.

Why 'we only serve lawful government' stopped being a defense

Palantir's standing answer to ethics critics has long been a version of: we build tools for the lawful institutions of democratic societies, and democracies decide their own policies at the ballot box, not at our keyboard. It's a coherent position. The problem is that the 2020 letter blew it up from the inside. By declining ERO and CBP work in 2020, Palantir conceded the principle it now disowns: that some lawful contracts carry human-rights risk grave enough to refuse, and that the company is the one who must do the refusing. You cannot say 'this is too dangerous to build' in 2020 and 'we just serve lawful government' in 2025 about the same category of work. One of those statements is being made in bad faith. The pension-fund trustees noticed.

The dissent arrived from inside the house, too - though not in the form commentary likes to claim. There was no ethics-board walkout. What is documented is narrower and, in a way, sharper: in May 2025, thirteen former Palantir employees, including software engineers and a member of the company's own privacy and civil liberties team, published an open letter condemning the ImmigrationOS contract and stating that Palantir's founding principles 'have now been violated.'⁶ Then in August 2025, Amnesty International - the same organization Palantir had written to in 2020 - assessed that ImmigrationOS's capabilities matched those of ICE's 'Catch and Revoke' system, and that Palantir had 'failed to fulfill its human rights responsibilities due to the lack of adequate human rights due diligence.'⁸ The 2020 promise and the 2025 indictment came from the same desk.

The fairest version of Palantir's case

Here is the honest counter, made as strongly as it deserves. First, a company is not a person, and a posture taken in 2020 is not a contract with the future; circumstances, administrations, and legal judgments change, and refusing to ever revisit a stance can itself be a failure of governance. Second, the 'surveillance company' label is genuinely overstated. Per Palantir's own FY2025 10-K, government was 54% of its $4.5 billion in revenue and commercial customers were 46% - a nearly even split, up from the 55/45 it reported a year earlier on $2.87 billion in revenue.⁴⁵ This is not a firm that lives or dies on ICE money; the deportation work is a thin, loud slice of a business increasingly funded by hospitals and manufacturers. Third, ImmigrationOS enforces laws passed by an elected government, and Palantir is right that it does not write those laws.

All true - and none of it touches the actual charge. The objection was never that Palantir is exclusively a surveillance firm, or that it can never change its mind. The objection is that it named a specific moral red line, in writing, to a human-rights organization - and then crossed that exact line without ever explaining what changed except the occupant of the White House. 'We revised our principles after careful review' is a defensible sentence. Palantir hasn't said it. What the record shows is a commitment made when declining was cheap, and abandoned when accepting became lucrative. That isn't a company evolving. That's a company discovering its principles were priced.

Palantir's defenders are right that it serves lawful government, that elections decide policy, and that it is not the author of the deportations its software accelerates. But it once stood somewhere more specific than that. It looked at the same work, named the same risks Amnesty later confirmed, and said no - on the record, to the world. The controversy isn't that Palantir builds for the state. It's that Palantir already told us which buildings it wouldn't enter, and then walked in, and never said why. A company that keeps its word doesn't need a third-party human-rights audit to defend it. A company that demonstrably broke its own gets one demanded by its shareholders - which is precisely what happened.