White Sands Missile Range sits on nearly 3,200 square miles of New Mexico desert, and keeping it running, tracking equipment, range safety, mission support, takes a contractor willing to sign up for years of unglamorous, expensive work. TRAX International Corp. had been doing some version of that job for the Army before. In 2026 it bid $420 million to keep doing it. It lost to a joint venture called Southwest Range Services, whose bid landed at $449 million, roughly $29.4 million higher, about 6.5 percent.
Losing a bid to a more expensive competitor happens. Federal evaluators are allowed to decide that a pricier proposal is technically stronger and worth the premium. What TRAX is now telling a federal judge is that the "stronger" finding against its own proposal was not real. According to the complaint, one of the weaknesses the Army assigned to TRAX's bid rested on "made-up references to TRAX's proposal, that no one on the [Source Selection Evaluation Board] checked." The filing calls it what it looks like: a classic AI hallucination, sitting inside a document that decided where nearly half a billion taxpayer dollars would go.
The Case, By The Numbers
TRAX first challenged the award at the Government Accountability Office, the normal first stop for a losing federal bidder. GAO denied that protest in May 2026. So TRAX went to the U.S. Court of Federal Claims instead, filing a sealed complaint that became public July 29. The case is docketed as TRAX Int'l Corp. v. United States, No. 26-cv-896, and assigned to Judge Carolyn N. Lerner. Southwest Range Services, the winning joint venture, is made up of Systems Application & Technologies Inc., Amentum and EWA Warrior Services, according to court filings reviewed by Bloomberg Law.
The complaint's central claim is narrow enough to test in court: that the Army's internal evaluation used an AI platform known as FAST TRACK during the review of proposals, and that FAST TRACK's output included evaluator-facing language about TRAX's bid that did not match anything actually in TRAX's proposal. Nobody caught it before the award went out. If a $29.4 million price gap turned partly on a weakness nobody can point to in the actual bid documents, TRAX argues, the award has to be redone.
What The Army Says Happened
The Army has not simply gone quiet on this. Court filings the government submitted at the end of July lay out its own account, arguing that FAST TRACK did not actually influence the award decision. That is the honest complication in this story: the Army is not conceding that an AI tool tainted a nearly half-billion-dollar procurement. It is disputing the premise while declining to discuss the litigation any further in public. Asked for comment on the record, the Army and the Justice Department both had nothing to add beyond what is already in the filings.
That leaves a federal judge to sort through something the government itself has not been fully transparent about in the past: which parts of a source selection came from a person on the evaluation board, and which parts came from software. Legal analysts who track federal procurement, including attorney David Timm, have pointed out that most agencies do not report AI use in bid evaluations as "high-impact" under existing disclosure rules, which makes it genuinely hard for a losing contractor to even prove an AI tool touched their file, let alone that it got something wrong.
Why This Is Bigger Than One Missile Range
The dollar figure is what got this story picked up. The mechanism is what should worry anyone who deals with the federal government. Source selection boards exist because picking a winning contractor is supposed to be a documented, defensible, human judgment call, one that a losing bidder can challenge and a court can review. If an AI tool is quietly doing part of that judgment call, and its output reads exactly like a human evaluator's notes, the paper trail stops meaning what it is supposed to mean. A "weakness" assigned to a bid is not just a data point. It is the reason a company loses a contract worth hundreds of millions of dollars, and until now nobody outside the Army had reason to ask whether a person actually wrote it down.
This case will not resolve quickly. TRAX is asking the court to order a new evaluation, not just a payout, which means a judge will need to dig into exactly how FAST TRACK was used and whether its output shaped what a human ultimately signed off on. That is a fact question, not a policy question, and fact questions in federal court take time. What is already on the record, regardless of how Judge Lerner rules, is that a defense contractor is now formally accusing the Army of letting an AI tool write parts of a source selection document that decided a $449 million contract, and that the Army's own evaluators apparently did not catch it.
The Verdict
TRAX International says an Army AI tool called FAST TRACK invented unsupported criticism of its bid, contributing to the loss of a $449 million contract at White Sands Missile Range to a proposal priced $29.4 million higher. The Army disputes that the tool influenced the outcome but has declined further comment. The case, now before the U.S. Court of Federal Claims, is the clearest public test yet of what happens when a government agency lets AI sit inside the paperwork that decides who gets paid.