The Pentagon formalises Palantir’s Maven AI as a core military system

The Pentagon intends to designate the Palantir Maven Smart System as an official program of record, locking in multi-year funding for the Artificial Intelligence (AI) enabled targeting platform that is already deployed across every US combatant command, according to a policy memo first reported by Reuters. Deputy Secretary of Defense Steve Feinberg signed the memo, stating that program-of-record status "will provide the stable funding and resourcing necessary" for Maven's continued development, integration, and use by commanders in combat operations, Bloomberg reported.

The designation enters Maven into the Future Years Defense Program as a protected line item, giving it visibility and stability across budget cycles that experimental programs lack. The US Army will manage all Maven contracts going forward, and oversight will transfer from the National Geospatial Intelligence Agency to the Chief Digital and AI Officer within 30 days, with program-of-record status expected before the close of fiscal year 2026 on September 30.

Maven started under Deputy Secretary of Defense Robert O. Work in April 2017 as an effort to apply machine learning to drone surveillance footage. Google was the original technology partner but withdrew in 2018 after employee protests.

Palantir took over and built a full command-and-control platform that ingests data from more than 150 sources, according to Palantir's public demonstrations: satellite imagery, drone video, radar, infrared sensors, signals intelligence, and geolocation data. Machine vision algorithms trained on millions of labelled images automatically detect and classify battlefield objects, with yellow-outlined boxes marking potential targets, blue outlines flagging friendly forces and no-strike zones, and an ‘AI Asset Tasking Recommender’ proposing which weapons platforms and munitions should be assigned to each target.

NGA Director Vice Admiral Frank Whitworth stated at Palantir's AIPCON 9 conference in March that Maven can generate 1,000 targeting recommendations per hour, as reported by The Register, with the 18th Airborne Corps reportedly achieving comparable targeting output to the 2,000-person cell used during Operation Iraqi Freedom with roughly 20 people. Maven now has more than 20,000 active users, a figure that has quadrupled since March 2024. The platform was used during the 2021 Kabul airlift, to supply target coordinates to Ukrainian forces in 2022, and most recently during Operation Epic Fury against Iran in 2026, where it reportedly enabled processing of 1,000 targets within the first 24 hours, according to SpaceNews. NATO acquired a version in March 2025.

The Future of AI Keynote: Dystopia, Utopia and Control, a keynote by futurist Matthew Griffin

An initial $480 million Maven contract was awarded in May 2024, with the ceiling raised to $1.3 billion in May 2025, alongside a $10 billion Army enterprise framework agreement signed in July 2025 that consolidated 75 existing Palantir contracts.

Meanwhile, the FY2026 defence budget reached $1.01 trillion, representing a 13% increase over FY2025, and for the first time included a dedicated AI and autonomy budget line of $13.4 billion, according to MeriTalk's analysis of the Pentagon budget request. That allocation covers unmanned aerial vehicles ($9.4 billion), maritime autonomous systems ($1.7 billion), and supporting AI software ($1.2 billion). The Pentagon now oversees more than 685 AI-related projects tied to weapons systems, per Congressional Research Service tracking.

Secretary of Defense Pete Hegseth's January 2026 AI strategy memo, published on defence.gov, declared the US military would become an "AI-first warfighting force" and outlined seven priority projects for FY2026, including Swarm Forge for autonomous drone swarms and Agent Network for AI-driven kill chain execution. The Drone Dominance Program aims to field more than 200,000 one-way attack drones by 2027, with a first order of 30,000 units at roughly $5,000 each, Breaking Defense reported.

Anthropic's Claude AI models were also integrated into Maven through Palantir's platform and received Impact Level 6 accreditation for classified environments. I’ve reported extensively on how Anthropic refused to allow Claude to power fully autonomous weapons or mass surveillance, with the Pentagon then designating the company a "supply chain risk to national security" in February. Anthropic filed federal lawsuits and OpenAI struck a deal with the Pentagon shortly after. Analysts estimate that extracting Claude from classified networks could take up to 18 months.

DoD Directive 3000.09, updated in January 2023, requires "appropriate levels of human judgment over the use of force" and senior-level review before autonomous weapons development. Hegseth's January 2026 strategy memo, however, pushes in the opposite direction, stating explicitly: "We must accept that the risks of not moving fast enough outweigh the risks of imperfect alignment,” with testing requirements treated as potential "blockers" subject to waiver by a newly created Barrier Removal Board.

The Brennan Center for Justice, in a March 2026 report titled "The Business of Military AI," documented that Hegseth halved staffing at the Office of the Director of Operational Test and Evaluation and shuttered the Civilian Protection Center of Excellence. The centre's researchers wrote that "the accelerating use of AI in warfighting has not been met with commensurate urgency to reckon with its dangers."

CSIS research has quantified AI-assisted targeting error propagation at 25% under variable conditions, according to a January 2026 analysis. Whitworth stated that by June 2026, Maven will begin transmitting "100 percent machine-generated" intelligence to combatant commanders. “No human hands actually participate in that particular template and that particular dissemination,” he added. “We want to use it for everything, not just targeting.”

Senator Elissa Slotkin introduced the AI Guardrails Act this month, which would prohibit the DoD from using autonomous weapons to kill without human authorisation and bar AI use for domestic mass surveillance, The Hill reported. The FY2026 NDAA already declares targeting and launch authorisation "inherently governmental" functions and requires reporting of autonomous weapons directive waivers to Congress.

The US isn’t alone in its push for AI in military applications. China's People's Liberation Army is pursuing "intelligentised warfare" under the 15th Five-Year Plan with a 2027 target for decisive military capability over Taiwan - a claim that US intelligence services have denied.

The Georgetown University Center for Security and Emerging Technology reviewed thousands of PLA procurement requests and identified AI decision-support systems, AI for countering US naval assets, and AI for information manipulation, according to a February 2026 report. Georgetown researchers noted no equivalent of Anthropic-style restrictions on autonomous kill chains within PLA procurement.

Meanwhile, a recent CSIS analysis documented Russian forces striking approximately 300 targets per day using unmanned systems in Ukraine, with data collection feeding AI platforms designated Platform-GNS and Avtomat. Russia voted against the December 2024 UN General Assembly draft resolution on lethal autonomous weapons alongside only North Korea and Belarus. That resolution passed 166-3 but remains non-binding; no international treaty currently governs lethal autonomous weapons systems. With AI reshaping the technology industry, its influence has now begun to slip into the long shadow of military usage, and the implications of such deals remains to be seen.