TRAX International Corp. is asking the U.S. Court of Federal Claims to order the Army to reevaluate bids for a mission-support contract at White Sands Missile Range, arguing in a lawsuit that an artificial-intelligence tool used during the source selection process "hallucinated multiple times" and cost the company a roughly $450 million award.
Background
TRAX International Corp. first challenged the award to Southwest Range Services, the incumbent contractor at White Sands, through a bid protest at the Government Accountability Office. GAO docketed the case as B-424271, B-424271.2 and B-424271.3, with TRAX arguing the Army mishandled an organizational conflict of interest, misapplied its mission-capability evaluation criteria and undervalued TRAX's past performance. GAO denied the protest on May 14, 2026, finding no reasonable possibility that any of the alleged errors changed the outcome of the competition. In its decision, GAO reiterated a standard the board applies broadly across protests: "Where the record establishes no reasonable possibility of prejudice, we will not sustain a protest even if defects in the procurement were found."
That denial did not end the fight. In late July 2026, TRAX filed a new action at the U.S. Court of Federal Claims, asking a judge to direct the Army to reevaluate the proposals. Unlike the GAO protest, the new complaint centers on a claim that did not feature prominently in the earlier filing: that the Army's evaluators relied on an artificial-intelligence tool that fabricated content used to justify the agency's source-selection decision.
Key Details
According to the complaint, the Army's AI evaluation tool generated a written "weakness" against TRAX's technical proposal that cited language TRAX says never appeared anywhere in its submission. TRAX's filing calls the finding a "classic AI hallucination," stating the weakness contained "made-up references to TRAX's proposal, that no one on the [Source Selection Evaluation Board] checked" before the board relied on it to help justify selecting Southwest Range Services.
The stakes attached to that single weakness are significant. Southwest Range's winning proposal was priced at roughly $449 million, a $29.4 million premium over TRAX's roughly $420 million bid, and the Army's source-selection authority cited technical superiority, including the disputed weakness, to justify paying that premium. In its earlier GAO decision, the agency's tradeoff rationale pointed to "the significantly greater importance the RFP placed on the mission capability factor, as well as the significantly greater importance of the non-cost/price factors when compared to cost/price" — the same evaluation record TRAX's new complaint says was tainted by fabricated AI content. TRAX's lawsuit argues that, "given the $29.4 million price premium" involved, "there is a reasonable possibility that the removal of this Weakness could have changed the award decision" — the legal threshold the Court of Federal Claims uses to determine whether a reevaluation is warranted.
The Army has declined to comment on the pending litigation. The dispute remains before the Court of Federal Claims, with no ruling yet on TRAX's request that the agency reevaluate the proposals.
TRAX's allegations are not the first time a contractor has accused a federal evaluation team of leaning on AI output that did not hold up to scrutiny. Salient CRGT raised similar claims against a Defense Department subagency, arguing that AI-assisted evaluation had produced unreliable findings against its proposal. GAO dismissed that protest as abandoned in a Jan. 5, 2026 decision, closing the matter without a ruling on the underlying AI allegations. That outcome left the merits of AI-reliability claims untested at GAO, making TRAX's Court of Federal Claims complaint a potential precedent for how federal judges evaluate claims that an agency's automated tools produced unreliable evaluation content.
The two cases point to a pattern rather than an isolated dispute. In both instances, the contractor's central complaint was not that an agency used AI in its evaluation process — procurement rules do not categorically bar that — but that the output was treated as reliable without adequate human verification before it became part of the official award justification. Neither TRAX nor Salient CRGT has publicly identified the specific AI system involved, and neither cited source describes documentation showing how the tools were validated before the evaluations at issue.
What It Means for Contractors
The White Sands dispute tests a growing problem: agencies incorporating AI tools into technical evaluations without consistently disclosing when or how the tools were used. TRAX's complaint hinges on the claim that no human on the Source Selection Evaluation Board caught the fabricated content before it was folded into the official record supporting the award. If that account holds up, it points to a verification gap rather than a policy gap — the Army may already require human review of AI-assisted findings, but the safeguard did not function as intended in this case.
For contractors preparing proposals against agencies known to be piloting AI evaluation tools, the practical lesson is to request debriefings that specifically address whether AI-generated content contributed to any assigned weaknesses or deficiencies, and to compare that content line-by-line against the submitted proposal. A weakness that cites language, page numbers or claims that do not appear anywhere in the actual submission is now a documented basis for both a GAO protest and, as TRAX's case shows, a follow-on Court of Federal Claims action if the protest is denied first.
The case also underscores that a GAO denial is not necessarily the final word. TRAX lost at GAO on prejudice grounds tied to its OCI, mission-capability and past-performance arguments, but it preserved a separate theory — one built specifically around the integrity of the AI-assisted evaluation record — for a second venue with its own standard of review. Contractors weighing whether to pursue litigation after a GAO loss should note that the Court of Federal Claims can revisit the administrative record independently, and that newly surfaced evidence of an unreliable automated process can support a fresh prejudice argument even after GAO has ruled.
Until the Court of Federal Claims rules in the case, the White Sands mission-support contract stays with Southwest Range Services under the original award. The litigation is already shaping how contractors weigh AI-related evaluation risk, and further protests raising hallucination claims look likely as more source-selection boards use AI review tools without uniform disclosure requirements.
