Re: Palantir Comments on Development of an Artificial Intelligence Action Plan
Published March 15, 2025 · Printed at the top of page 1, below the public-dissemination notice and above the addressee (Mr. Faisal D’Souza, National Coordination Office). The PDF metadata gives a creation date of 14 March 2025, one day earlier, consistent with a memo finalised for the 15 March 2025 comment deadline.
Not law. This is a company's own public position on AI regulation. It is not law, and it carries no legal force.
What it argues for
This is Palantir Technologies' submission to the White House's February 2025 request for information on an AI Action Plan (Federal Register document 2025-02305), a memo addressed to the National Coordination Office and published as a PDF on palantir.com. It frames AI policy as a contest with China — "We should meet this challenge as Americans always have: with bravery, tenacity, and respect for the enormous responsibilities of freedom" — and deliberately confines itself: "we have focused our recommendations on the software and application portions of AI policy". Its recommendations fall under five headings. First, procurement: "the federal government must streamline and accelerate the acquisition and implementation of AI capabilities across federal agencies", which means that "Complying with FASA should include prioritizing procurement of modifiable commercial products over custom development, absent rare and justifiable circumstances", reviewing in-flight projects for compliance, reforming the Authority to Operate process and expanding Other Transaction Authorities with fixed-price and standard commercial IP terms. Second, data infrastructure: dedicated testing and evaluation capability that can "validate performance prior to general release", and privacy and data governance, since "Privacy is not just a matter of individual rights; it is a national security issue." Third, incentives: agency AI testbeds using "operationally relevant data (in appropriately deidentified or minimized form)" and a goal "that 1% of every agency budget be spent on AI modernization efforts". Fourth, quick wins in military command and control, healthcare data and fraud detection, plus "a mandate for each government agency to complete a new flagship AI project within nine months of the Action Plan being published". Fifth, global leadership: "the United States should host the next global AI summit", in Silicon Valley, and export controls "to prevent the most advanced AI chips from reaching China", backed by software to detect smuggling. It closes that "The AI race is as much a competition for the elevation of values and first principles as it is a critical contest for technological supremacy."
Stated positions (14)
- Federal AI acquisition must be sped up: "the federal government must streamline and accelerate the acquisition and implementation of AI capabilities across federal agencies".
- Commercial products should beat custom builds by default: "Complying with FASA should include prioritizing procurement of modifiable commercial products over custom development, absent rare and justifiable circumstances", and "it would be prudent to review all proposed and inflight projects for FASA compliance".
- Security accreditation should be faster and shared between agencies: "The ATO process for AI systems should be streamlined and accelerated", with a risk-based approach and "a clear path to reciprocity between agencies".
- OTAs for AI should carry commercial terms, potentially "mandating that firm, fixed-price contract types and standard commercial intellectual property terms are included in such agreements to prevent the trend of turning OTAs into FAR-like vehicles".
- Government needs dedicated testing capacity: "Testing and Evaluation (“T&E”) is critical for getting AI from prototype to production", and "the Administration should prioritize establishing dedicated T&E capabilities for AI systems that are able to validate performance prior to general release and assist with ongoing monitoring to track performance in operational use".
- Privacy is framed as security: "Privacy is not just a matter of individual rights; it is a national security issue", with access controls, data deletion, data minimisation and "comprehensive data lineage tracking to maintain transparency and accountability in AI systems".
- Agencies should open their data to developers through testbeds, "creating agency-specific AI testbeds that utilize operationally relevant data (in appropriately deidentified or minimized form)", with agency privacy officers approving de-identification.
- A spending floor for AI: "the Trump Administration should set as a goal that 1% of every agency budget be spent on AI modernization efforts."
- Every agency should ship an AI project on a deadline: the Action Plan should include "a mandate for each government agency to complete a new flagship AI project within nine months of the Action Plan being published".
- Quick wins named in defence, health and fraud: "Expand funding for integration of AI into C2 systems", expand healthcare data investment, and use AI for fraud because "Procurement fraud at the DoD, through efforts like inflated billing or rigged bidding, could be more easily detected using AI-driven models".
- The US should host the next AI summit: "the United States should host the next global AI summit", and "The next summit should be held in Silicon Valley."
- Chip export controls should be tight and enforced: "the United States must continue to implement and enforce robust export control policies to prevent the most advanced AI chips from reaching China", and "These policies should include strict regulations on the export of high-end AI chips, regular updates to the list of controlled technologies, and collaboration with allied nations to create a unified approach."
- Enforcement should use software: "it is essential to develop and deploy advanced software technology capable of detecting and preventing smuggling through third-party countries and companies".
- AI competition is framed as a contest of values with allies: "The AI race is as much a competition for the elevation of values and first principles as it is a critical contest for technological supremacy", working through the Quad, NATO, the G7, IPEF, AUKUS and the Five Eyes.
About this document
A 13-page PDF memo in letter form on palantir.com, headed "Re: Palantir Comments on Development of an Artificial Intelligence Action Plan", addressed "Dear Mr. D’Souza" at the National Coordination Office and marked as an email submission to [email protected] under Document Number 2025-02305. It opens with a notice that it is approved for public dissemination and may be reused by the government "without attribution". It is written in the name of Palantir Technologies Inc.; there is no signature block or individual byline, and no footnotes. After a one-page introduction and a page listing five mission sets, a "RECOMMENDATIONS" section gives numbered recommendations with lettered sub-recommendations: 1 Modernize Federal AI Acquisition and Deployment (FASA, ATO reform, OTAs); 2 Invest in Foundational Data Infrastructure (T&E; privacy, security and data governance); 3 Initiatives and Incentives (testbeds; 1% of agency budgets); 4 Existing Opportunities (C2, healthcare data, fraud detection, flagship projects); 5 Maintain American Global Leadership (a Silicon Valley summit, chip export controls, values and alliances). It ends with a one-page "CONCLUSION". It cites Secretary Hegseth's software-acquisition memo, Senator Wicker's FoRGED Act, Project Maven and the National COVID Cohort Collaborative.
How this sits against AI law
Each stance compared with what EU and US instruments actually require. Where no instrument addresses a theme, that gap is shown rather than hidden.
Commercial-first federal procurement of AI
Agencies should comply with FASA by buying modifiable commercial products over custom development except in rare cases, review in-flight projects for compliance, streamline and make reciprocal the Authority to Operate process, and expand Other Transaction Authorities with firm fixed-price contracts and standard commercial IP terms.
The AI Act regulates providers and deployers of AI systems, not how public bodies procure them; it contains no preference for commercial products and no procurement procedure.
OMB M-25-22 sets a lifecycle process for federal AI acquisition and favours a competitive market and US-developed AI, but it also requires contract terms on government data rights, portability, vendor lock-in and limits on training commercial models with non-public government data — protections that pull against Palantir's ask for standard commercial IP terms.
Dedicated testing and evaluation before release and monitoring in use
The government should build dedicated T&E capability to validate AI performance before general release and monitor it in operation, set clear fielding-certification guidelines, and train small government T&E teams to work alongside vendors.
For high-risk AI systems the Act requires a risk-management system that includes testing before the system is placed on the market, and a post-market monitoring system once it is in use; these duties fall on providers rather than on a government testing body.
The Action Plan's "Build an AI Evaluations Ecosystem" section asks NIST and CAISI to publish guidelines and resources for federal agencies to conduct their own evaluations of AI systems, supports the science of AI measurement and funds AI testbeds through DOE and NSF.
Privacy, security and data governance as national security
Privacy is a national security issue; federal AI programmes should rest on robust access controls, efficient data deletion, data minimisation and comprehensive data lineage tracking, so that systems can be tested, overseen and audited.
The Act sets data and data-governance requirements for the training, validation and testing data of high-risk systems and requires automatic record-keeping (logging) for traceability, while the GDPR continues to apply to personal data alongside it.
M-25-22 requires cross-functional acquisition teams that include privacy and security staff, and contract terms addressing rights in government data and restrictions on using non-public government data to train commercial AI systems.
Opening government data to AI developers through agency testbeds
Each agency should create an AI testbed that gives commercial AI companies and researchers secure access to operationally relevant government data, de-identified or minimised and approved by agency privacy officers, with clear IP frameworks and feedback loops.
The Act provides for AI regulatory sandboxes, which the 2026 Digital Omnibus on AI (RAI-EU-NA-COM2025-2025) expands, but these are supervised environments for developing and testing AI against the law, not a programme that opens public bodies' operational data to industry.
The March 2026 legislative recommendations say "Congress should provide resources to make federal datasets accessible to industry and academia in AI-ready formats for use in training AI models and systems", and ask Congress to establish regulatory sandboxes for AI applications.
A 1% budget floor and flagship AI projects in every agency
The Administration should set a goal that 1% of every agency budget be spent on AI modernisation, and the Action Plan should require every agency to complete a new flagship AI project within nine months, approved through OSTP.
The AI Act sets no spending targets or adoption mandates for public bodies.
OMB M-25-21 requires agencies to designate Chief AI Officers and publish AI strategies to accelerate adoption, but sets no spending floor and no project deadline; the Action Plan's government-adoption section did not adopt either proposal.
AI in defence command and control
Funding for integrating AI into military command-and-control systems should be expanded, with clear pathways for moving AI experiments into production, building on programmes such as Project Maven.
The AI Act does not apply to AI systems placed on the market or used exclusively for military, defence or national-security purposes.
The Action Plan has a section, "Drive Adoption of AI within the Department of Defense", including a streamlined DOD process for classifying, evaluating and optimising workflows for AI and priority agreements with cloud and compute providers for use in a national emergency.
Export controls on advanced AI chips and anti-smuggling enforcement
The US must enforce robust export controls to keep the most advanced AI chips from China, with strict rules on high-end chips, regular updates to the controlled list and a unified approach with allies, enforced with software that detects smuggling through third countries.
The EU's dual-use regulation controls exports of listed dual-use goods, software and technology and its control list is updated periodically (most recently by Delegated Regulation (EU) 2025/2003), but the EU has no AI-chip regime targeted at China of the kind Palantir describes.
The Action Plan's "Strengthen AI Compute Export Control Enforcement" section proposes exploring "location verification features on advanced AI compute to ensure that the chips are not in countries of concern" and a DOC effort with the intelligence community on global chip export-control enforcement; the underlying controls sit in the Export Administration Regulations (RAI-US-NA-EARAICN-2023).
US leadership of global AI governance through summits and alliances
The United States should host the next global AI summit, in Silicon Valley, and lead AI development with like-minded partners through the Quad, NATO, the G7, IPEF, AUKUS and the Five Eyes, as a contest of values against authoritarian adversaries.
The EU's values-based international commitment is its signature of the Council of Europe Framework Convention on AI, a human-rights treaty open to non-European states; it is not a US-led alliance and says nothing about hosting summits.
The Action Plan's international pillar concentrates on exporting the American AI stack to allies and on export controls, not on hosting a summit.
Much of this memo reads as a draft of what the White House then published: America's AI Action Plan (July 2025) has the evaluations ecosystem and AI testbeds Palantir asked for, a GSA-run AI procurement toolbox, a push for AI adoption in the Department of Defense and a section on chip export-control enforcement that proposes exploring location-verification features on advanced chips, much as Palantir proposed software to detect smuggling. Palantir went further on money and deadlines — a floor of 1% of every agency budget and a nine-month flagship-project mandate — which the Action Plan did not adopt, and on commercial terms: OMB's April 2025 acquisition memo M-25-22 pulls the other way by requiring contract terms that protect government data and intellectual-property rights rather than "standard commercial intellectual property terms". Against the EU the contrast is structural: the AI Act does not regulate how governments procure AI and excludes military and national-security uses altogether, which covers the defence use cases (command and control) Palantir foregrounds; where the two do overlap — testing before release, post-market monitoring, data governance, sandboxes — the EU already makes for high-risk systems binding what Palantir asks the US government to build for its own use.
Source
https://www.palantir.com/ai-action-plan-recommendations.pdf- Date on the page:
- March 15, 2025
- Source checked:
- opened and confirmed on 2026-09-30