“Operation Epic Fury leveraged Palantir’s Maven Smart System in order to conduct strike missions across the entire battle space, 13,000 targets in 38 days,” Cameron Stanley, the Pentagon’s Chief Digital & AI Officer (CDAO), told the SCSP AI+Expo — a claim that, if taken at face value, captures both scale and a change in how the Defense Department used artificial intelligence during the recent campaign with Iran.
Maven Smart System: from Project Maven to a live battlespace tool
Maven Smart System (MSS) began as an evolution of Project Maven and now exists as a multi-purpose military planning suite built by Palantir, with a separate offshoot run by the National Geospatial Intelligence Agency. Palantir describes the toolkit as “a live, synchronized view of the battlespace” that “enables synchronized mission planning and execution,” and lists features such as “automated object detection” and “centralized target identification and management.” According to Pentagon officials cited in public reporting, MSS also analyzes combinations of video, satellite imagery, and other technical intelligence sources and can sift through extensive libraries of reports to draft summaries or propose potential courses of action.Scale and the spike in usage during Operation Epic Fury
Pentagon officials and newly disclosed usage figures indicate a sharp rise in MSS activity during the 38-day air campaign identified as Operation Epic Fury. A Pentagon spokesperson told Breaking Defense that unclassified usage rose 38 percent month-to-month and classified usage rose 89 percent over the same interval. Measured by “tokens” — the mathematical operations that underlie generative AI models — peak daily usage climbed 4,425 percent; at one point daily token consumption reportedly reached approximately 20 billion tokens. For context provided in the reporting, civilian interactions with chatbots typically expend several hundred tokens per question-and-answer, while an individual power user in a high-end paid account might use up to a quarter-million tokens in a single day.Tools and users: low-code interfaces and semi-autonomous agents
Officials said MSS now includes “low-code/no-code” tools intended to let less technical users build software and even semi-autonomous AI “agents” to carry out routine tasks. That capability aligns with Stanley’s assertion that troops have shown an “insatiable appetite” for AI tools that “allow us to take all of this data, synthesize the data, and make better decisions, faster, on the battlefield.” The Pentagon declined to provide public examples of MSS outputs from the Iran strikes, but spokespeople and contractor materials indicate the system is being used to streamline planning, target identification, and the synthesis of intelligence products.Compute, capacity and the CDAO’s central worry
Stanley framed the broader logistical challenge plainly: the Department is consuming ever more compute to run these systems, and his “biggest fear” is whether the Pentagon can keep up. He said the Department is exploring ways to increase capacity “in every domain, in every classification level,” from junior operators to senior commanders, to satisfy that demand. The rising token counts and dramatic month-to-month usage spikes underscore the scale of the capacity question: sustaining tens of billions of tokens a day represents a materially different resource profile than the modest token counts associated with typical civilian chatbot use.Risks, testing, and the human-machine balance
Stanley acknowledged risks tied to accelerating AI into operational decision-making and said the CDAO office is “diligently testing its AI algorithms.” He emphasized a model of human-machine teaming in which technology processes data at speed while humans “analyze the situation, apply operational art, context, and legal intent,” to reduce errors and secure “decision superiority on the battlefield.” He did not, however, offer public examples of MSS-related collateral effects from the Iran strikes. The reporting notes that the campaign’s first day “killed at least 175 at a school adjoining an Iranian Revolutionary Guard Corps base,” reportedly due to out-of-date intelligence, and adds that such deadly errors have occurred in pre-AI eras as well.What this means for technologists, policymakers, and commanders
- Technologists and security teams — Will need to plan for dramatically larger compute footprints and expanded testing regimens as MSS-style tools scale beyond experimental use, a point the CDAO singled out when he asked whether the Department can “keep up.”
- Policymakers and acquisition officials — Face decisions about provisioning compute capacity “in every domain, in every classification level,” and about governance frameworks for low-code tools and semi-autonomous agents deployed to non-expert users.
- Operational commanders and operators — Have already shown strong demand: the CDAO described an “insatiable appetite” among troops for AI capabilities that accelerate data synthesis and decision timelines, even as leaders stress continued human judgment to check machine outputs.
