Pentagon's Maven Smart System Tops 100,000 Users as Verification Lags Behind
Maven Smart System use across the U.S. military has passed 100,000 personnel, up from roughly 50,000 in January, according to James Mazol, a deputy director at the Department of Defense. Mazol linked the increase to the demands of the Iran conflict and to adoption spreading through the combatant commands and the National Guard Bureau. The Palantir-built platform was used to work 13,000 targets across 38 days of Operation Epic Fury, with about 1,000 struck on the opening day. The disclosure lands weeks after a Pentagon review tied overreliance on the same system to a strike that killed more than 120 civilians.
That pairing is the story. The user curve is the clearest available measure of how quickly AI has moved into the U.S. targeting workflow, and it is climbing faster than the verification machinery around it. Mazol's figure is not a pilot metric or a research benchmark. It counts people with access to a tool that fuses sensor feeds, satellite imagery and intelligence streams into target recommendations for commanders, compressing work that once consumed hours or days into minutes.
What the Maven Smart System growth curve measures
The headline number is 100,000, but the underlying series is messier than a single doubling. Palantir has claimed the Maven user base doubled every six months for more than two years. In May 2025, public estimates put it above 20,000 users, which implied a base near 80,000 by mid-2026 had the rate held. Mazol's January baseline of 50,000 sits above that implied trajectory, and the September figure of 100,000 sits above the projection again.
Some of the gap is definitional. Maven has been counted in terms of named accounts, active users and deployed seats, and those categories produce different totals. Reporting in early 2026 put the system at more than 20,000 active users spread across 35 service and combatant command software tools and three security classification domains. A separate count placed it above 25,000 accounts across every combatant command. Neither reconciles cleanly with 50,000 in January.
The operational tempo explains the demand. During the 38-day campaign, the system supported roughly 1,000 strikes in the first 24 hours, a volume officials described as unprecedented. That figure reframes what Maven is. It is not a decision-support tool pulled out occasionally for high-value targets; it is the throughput layer that made a strike rate of that scale administratively possible. Growth in users follows from that throughput rather than standing as a separate trend.
| Metric | Figure |
|---|---|
| Maven users, January 2026 | ~50,000 |
| Maven users, September 2026 | more than 100,000 |
| Targets worked, Operation Epic Fury | 13,000 over 38 days |
| Targets struck on opening day | ~1,000 |
| Contract ceiling | above $1 billion |
| Civilian deaths cited in Pentagon review | more than 120 |
For decision-makers, direction matters more than the absolute count. Every version of the series points the same way: steep, sustained, multi-year growth in the number of people whose work is mediated by an AI targeting layer. Program-of-record status locks that in.
The contract, the arsenal and the institutional bet
Maven carries a contract ceiling above $1 billion, with at least one accounting putting the value at $1.3 billion through 2029. The Pentagon is moving the platform to formal program of record status by the end of the fiscal year, a designation that converts an experimental capability into a budget line with its own requirements and oversight chain.
The department is also seeking funding for an AI Arsenal: sovereign, secure data centers built to train and run classified AI models. Read together, the program-of-record transition and the data center request describe a deliberate choice to own the infrastructure layer rather than rent it. That reduces dependence on commercial cloud regions for sensitive workloads and raises the fixed cost of the enterprise. Once sovereign capacity exists, the institutional incentive is to keep it busy.
The lineage reinforces how durable the bet is. Project Maven began in 2017 as the Algorithmic Warfare Cross Functional Team, with the Pentagon contracting Google to build AI for intelligence workflows. Palantir later became the platform provider. The system now runs on commercial foundation models, with Anthropic's Claude among those identified in accounts of the deployment. What began as an experiment in object detection is now infrastructure that most of the force touches.
Expansion beyond the combatant commands matters for the same reason. Bringing the National Guard Bureau and domestic-facing commands onto a targeting platform widens the set of missions Maven is expected to support, from overseas strike planning to disaster response and homeland security workflows. Each new mission type adds a governance question the original intelligence workflow never had to answer.
The verification gap
The Minab school strike is where the adoption curve and the accountability curve separate. A Pentagon review attributed the deaths of more than 120 civilians to overreliance on Maven, and the response has been to modify the system rather than pause it. Cameron Stanley, who heads the Pentagon's digital and artificial intelligence office, has described dozens of upgrades, including open-source feeds that track civilian maritime traffic, a capability that became more relevant as the conflict progressed.
Those changes are real and they are incremental. Adding a data feed improves what the model can see. It does not settle how much weight a commander should place on a machine-generated recommendation when the decision clock has been compressed from days to minutes. Speed is the platform's core value proposition, and speed is precisely what makes independent verification harder to insert.
There is also a measurement problem underneath the accountability one. The Pentagon publishes adoption figures with precision and accuracy figures not at all. A user count is easy to instrument. A false-positive rate on target recommendations requires ground-truth data that only becomes available after a strike, and that data is often contested and rarely released. The Minab review is one of the few public attempts to close the loop, and it arrived months after the strike.
The incentive structure points the same way. Officials have framed Maven's expansion as proof that it has graduated from pilot projects into everyday operational planning, and the department has set a goal of becoming an AI-first organization. In that frame, more users equals success. The civilian casualty review becomes a defect to be patched rather than a signal to slow down.
Competitors face the same governance problem from the other direction. Any vendor selling AI into the kill chain now inherits the Minab precedent: capability claims will be tested against civilian casualty outcomes, and the burden of proof shifts toward demonstrable accuracy rather than raw throughput. For Palantir, program-of-record status is a commercial win and a long-term liability exposure at once, because it attaches a named company to a named incident in a way that pilot programs never do.
The tension is not unique to Maven. Israel's military deepened its reliance on commercial AI infrastructure sharply in the years before 2024, and the pattern across modern militaries repeats: adopt the tool, scale the user base, retrofit governance afterwards. The question for the Pentagon is whether the retrofit keeps pace with a user base that doubled in eight months.
Why this matters
Maven's trajectory is the most concrete evidence yet that AI-assisted targeting has crossed from experiment into standard practice, and program-of-record status will make that harder to reverse. For defense planners and the vendors serving them, the near-term demand signal is clear: sovereign compute, model integration and the governance layer that decides when a human must override the machine. The risk is just as clear. A system that doubled its user base in eight months while a Pentagon review blamed it for more than 120 civilian deaths has not shown that its verification capacity is growing at the same rate. The metric to watch is whether the Pentagon publishes any independent measure of targeting accuracy alongside its adoption figures. Adoption on its own will keep rising.
Related Articles
- CENTCOM Patches the Maven Smart System Instead of Retiring It After Minab [Update]
- Palantir's Maven Program of Record Status Locks In $2.3B in Pentagon Funding
- ChatGPT Mil and Grok on GenAI.mil Now Reach 3 Million Users
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.