Maven and USAi: What Mature Federal AI Acquisition Actually Looks Like
Most of GAO’s April 2026 report on federal AI acquisitions (GAO-26-107859) documents failure modes—programs that didn’t document lessons learned, contracts that lacked AI-specific terms, programs retired without institutional postmortems. Two acquisitions stand apart as comparative benchmarks: DOD’s Maven program and GSA’s USAi platform. The report uses them to illustrate what AI acquisition looks like when agencies have had time to learn from their own mistakes.
Maven
Project Maven is DOD’s longest-running high-profile AI acquisition. Managed by the National Geospatial-Intelligence Agency, it uses machine learning and computer vision to analyze geospatial imagery and identify potential targets for human assessment. It has had a complicated history—including a period of public controversy over its AI ethics implications—but from an acquisition standpoint it represents an accumulation of hard-won institutional knowledge.
Maven’s early contracts were not well-specified. NGA officials told GAO that DOD initially funded the program without well-defined or documented requirements in order to acquire capabilities as fast as possible. The practical consequence: without clear requirements embedded in contracts, they couldn’t hold vendors accountable for the weekly deliverables needed under Agile development. That failure became the foundation of subsequent improvement. NGA progressively added more specific AI-related requirements in follow-on contracts. It developed cross-functional teams spanning computer scientists, software engineers, test and evaluation officials, cybersecurity experts, and product managers. It convened a two-day summit specifically on data rights and IP contract language. It awarded small contracts to multiple vendors to test their solutions before committing to larger acquisitions. And it developed comprehensive cost estimates that account for sustained infrastructure requirements—not just model training costs—to prevent the total cost of ownership surprises that trip up less mature programs.
The Maven approach to IP is particularly notable given how often data rights failures appear elsewhere in the report. Officials came to treat data ownership as the central strategic concern of AI acquisition, not the algorithm itself—a formulation that appears repeatedly among the report’s more experienced acquisition officials.
USAi
GSA’s USAi platform provides generative AI chatbot capabilities to federal agencies government-wide. Where Maven represents maturity earned through early failure, USAi represents an organization deliberately transferring lessons from prior AI work into a new acquisition.
GSA officials told GAO they reused contract terms that had proven effective in earlier AI acquisitions rather than building new frameworks from scratch—a form of institutional knowledge application that is only possible if the earlier work was documented. On data ownership, GSA developed a USAi-specific privacy policy and contract language clearly delimiting data ownership expectations and restricting vendor access to chat interaction data.
On pricing, USAi adopted a usage-based model rather than licensing—explicitly to avoid the situation where agencies pay for broad licenses that most employees will rarely use. GSA also built in structured pricing that accounts for the complexity of expected model usage, so that simple prompts are priced differently from sophisticated model features. This is a response to the pricing opacity problem that GAO found at every agency: by forcing pricing transparency into contract structure from the outset, USAi reduces the risk of being held hostage to vendor pricing decisions post-award.
For testing, USAi runs a battery of reliability evaluations against competing AI models over time. The explicit goal is to determine whether new, more expensive model versions are actually superior—providing the evidentiary basis for or against price increases that agencies otherwise have no way to evaluate.
GSA’s partnerships with Anthropic, OpenAI, and Google under the broader OneGov Strategy—formalized in August 2025—bring the USAi model to a government-wide scale. Whether that scale multiplies the benefits or introduces new vendor dependency risks is a question the report leaves open.
What These Cases Teach
The common thread in both Maven and USAi is that acquisition maturity is not a function of budget size or technical sophistication—it’s a function of institutional memory. Both programs applied lessons from prior failures, documented what worked, and built those findings into subsequent contracts. Both developed cross-functional teams that gave procurement officials access to the technical expertise needed to make good decisions. Both took pricing and data rights seriously as structural problems requiring contractual solutions, not just negotiating positions.
The report’s core argument is that these outcomes should not be exceptional. They should be the baseline—and the gap between Maven/USAi and the programs that retired without postmortems is precisely the gap that systematic lessons-learned collection is meant to close.