TRAX Files an AI Hallucination Bid Protest Over the $449 Million White Sands Award

TRAX’s AI Hallucination Bid Protest Centers on a $449 Million Recompete

TRAX International Corporation has filed an AI hallucination bid protest at the Court of Federal Claims (COFC). TRAX’s late-July 2026 complaint says the Army’s FAST TRACK tool hallucinated a technical weakness in TRAX’s proposal. That weakness tipped a $29.4 million premium toward Southwest Range Services, LLC, which won the $449.37 million White Sands Missile Range recompete. TRAX bid lower, at $419.99 million, and still lost. GAO denied TRAX’s earlier protest of the same award on May 14, 2026. That decision never reached the AI question. If you hold a recompete at an agency piloting an internal AI evaluation tool, treat this case as your playbook. I walk through what to document below.

Three facts anchor this story:

• The Army’s Source Selection Support Center of Excellence adopted FAST TRACK to speed up evaluations, according to Army’s own published account.
• GAO denied TRAX’s underlying protest on May 14, 2026 (B-424271, B-424271.2, B-424271.3), even though the Army conceded one evaluation finding was erroneous.
• TRAX’s COFC complaint, filed roughly two months after that denial, is the first place the AI-causation allegation appears in the public record.

The Army Conceded the Weakness Was Wrong. GAO Still Denied the Protest.

TRAX’s technical proposal drew a continuity-of-operations weakness during evaluation. The Army later conceded that finding was erroneous, according to GAO’s decision. GAO denied the protest anyway. Its reasoning is narrow: the source selection authority never cited that weakness as a basis for the award. No citation, no prejudice, per GAO’s test. That is a defensible legal standard. It is also why a conceded AI-generated error can survive a bid protest untouched.

Why GAO’s Bid-Protest Standard Never Reached the AI Hallucination Question

TRAX’s AI hallucination bid protest turns on a standard GAO applies to every case: did the error prejudice the outcome? It doesn’t ask what caused the error. A hallucinated finding and a human typo get the same treatment, as long as neither one drove the award decision. That’s the gap TRAX is now trying to close at COFC. COFC is a different venue, with different remedies, and no harmless-error shortcut built the same way.

Best-value tradeoffs like this one are ordinary. Agencies pay more for a technically superior proposal all the time. What makes this case different: the technical edge that justified the $29.4 million premium rested partly on a finding the Army already conceded was wrong. TRAX’s sharper claim is that an AI tool, not a human evaluator, generated that finding. GAO’s own record never had to test it.

FAST TRACK Is Real, and It’s Expanding Beyond Where It Started

FAST TRACK isn’t a rumor or a contractor’s guess about internal Army tooling. Army’s own published account describes the Source Selection Support Center of Excellence (S3COE) adopting it to speed up source selection. That account also shows the tool moving beyond its original home at ACC-Detroit Arsenal to ACC-Rock Island. This is AI in government source selection, already in production. It has outgrown the one-off pilot confined to a single contracting shop.

That expansion runs alongside a federal AI-governance regime. It took effect a year before TRAX’s award. OMB Memorandum M-25-21, issued April 3, 2025, requires agencies to inventory AI use cases. Agencies must also label qualifying ones “high-impact,” and that label triggers disclosure duties. Whether FAST TRACK carries that label isn’t established.

• “High-impact” under M-25-21 turns on whether AI drives a decision with a significant effect on someone, not just whether an agency uses AI at all.
• A tool that only assists a human evaluator, without deciding on its own, could plausibly fall outside that threshold. That’s the counterargument this gap doesn’t rule out.
• Even so, I found no public FAST TRACK disclosure addressing the question either way, despite Army’s own account of the tool’s active, expanding use.

This Isn’t the First Time: Salient CRGT Raised a Similar Claim in January

TRAX isn’t the first contractor to argue an AI tool produced a flawed federal evaluation finding. GAO dismissed a similar Salient CRGT allegation on January 5, 2026, per trade-press reporting. Two cases in eight months don’t establish a trend. They establish a track record. GAO’s harmless-error standard has already absorbed one version of this argument, without ever resolving who, or what, produced the finding.

Two independent oversight regimes sit side by side here: GAO’s protest-review standard, and OMB’s AI-disclosure requirement. Neither forces a “human or AI” answer into the record before an award survives review. My own inference is that GAO’s process is structurally unequipped to catch this pattern. That rests on two cases, not a policy concession. Treat it as my working thesis, not a settled finding.

TRAX and Southwest Range Services Are the Only Two Named Competitors in This Niche

Two named companies compete in this niche. TRAX International Corporation, based in Las Vegas, Nevada, describes itself on its own contracts page as the WSMR incumbent since 1979. I haven’t cross-checked that claim against a Tier 1 FPDS predecessor-contract record. Southwest Range Services, LLC, based in Las Cruces, New Mexico, is the awardee on RFP W51EW725RA003.

I found no public source naming other range and test-support-services competitors beyond these two parties. For firms in adjacent test-range categories at other DoD facilities, this is a two-party dispute. It says nothing about the broader competitive field.

What to Do Before Your Next Recompete Evaluation

The stakes aren’t hypothetical for anyone holding a federal services contract due for recompete. An unverified, AI-generated finding can cost you the award even when it’s wrong. GAO’s own standard isn’t guaranteed to catch it in time.

TRAX’s own timeline is the model. Three clocks run after an unfavorable technical evaluation. They don’t run concurrently:

1. Your debrief window opens first, usually just a few business days. Everything after it depends on the evaluation record you get back.
2. Your GAO protest clock starts next: generally 10 days from when you knew or should have known your protest ground (4 C.F.R. § 21.2).
3. Your COFC clock, under 28 U.S.C. § 1491(b), runs separately and longer. TRAX used that venue roughly two months after its GAO denial. It doesn’t restart when GAO rules.

Different readers carry different exposure here. The next move isn’t the same for everyone:

• BD leads and capture leads at DIB primes and DIB subs (often the founder or CEO directly, at smaller firms) undergoing a recompete: request the full evaluation record and ask whether AI-assisted tools contributed strengths or weaknesses before accepting an unfavorable result.
• Legal and GC counsel handling protests: treat an unexplained or unverifiable specific-language weakness as a potential AI-hallucination signal worth investigating before treating a GAO denial as final. COFC may be a necessary second-stage venue, because GAO’s harmless-error standard doesn’t probe origin.
• Agency-side contracting officers and program officers: confirm whether your office’s AI-assisted evaluation tools have been assessed and labeled “high-impact” under OMB M-25-21 before your next award cycle.

• You have your firm’s written debrief and the underlying evaluation record on file, not just a verbal summary.
• You’ve identified the specific evaluation factor that cost you the award, and checked its language against your own proposal’s actual wording.
• You’ve confirmed whether your GAO 10-day filing window is still open, or whether COFC is now your only remaining venue.

FAQ

What did the Army’s FAST TRACK AI tool actually do in the White Sands Missile Range evaluation, and did it cause TRAX’s protest-losing weakness?

FAST TRACK’s specific role in the WSMR evaluation isn’t established in the public record. TRAX’s COFC complaint alleges the tool generated the continuity-of-operations weakness the Army later conceded was erroneous. That is TRAX’s own unadjudicated claim, not a GAO or COFC finding.

Which other ACC contracting shops does S3COE support besides Detroit Arsenal and Rock Island?

Army’s own account names only ACC-Detroit Arsenal and ACC-Rock Island as confirmed users. I found no public list of every shop S3COE supports. Treat the tool’s reach as at least these two units, not necessarily limited to them.

Is this a preview of what other DIB incumbents should expect as agencies adopt tools like FAST TRACK?

It’s an early data point, not a confirmed pattern. FAST TRACK’s expansion to ACC-Rock Island is one signal. Salient CRGT’s earlier claim against a different evaluation is another. Together they suggest other DoD components piloting comparable tools carry similar exposure. No broader, multi-agency pattern is established yet.

What does OMB’s “high-impact” AI disclosure requirement actually require agencies to say about tools like FAST TRACK, and why isn’t that showing up in practice?

OMB Memorandum M-25-21 requires agencies to inventory their AI use cases. Any use case that meets the “high-impact” threshold then carries specific risk-management and disclosure obligations. I’ve found no public disclosure meeting that standard for FAST TRACK, despite Army’s own account of the tool’s active use. A tool that only assists a human evaluator, rather than deciding on its own, could plausibly fall outside that threshold in the first place.

What should I do if I lose a recompete and suspect an AI tool contributed to the evaluation against me?

Request your full evaluation record from the contracting officer, not just the debrief summary. Ask directly whether AI-assisted tools contributed any finding. Do this before your GAO protest window closes. COFC’s longer statute of limitations gives you a second venue if new evidence surfaces later, the way it did for TRAX.

Could TRAX still win, and what happens to the award if it does?

TRAX’s odds aren’t established in the public record. If TRAX prevails at COFC, the likely remedies are a corrective re-evaluation or a new award decision, not a guaranteed contract. That’s how bid-protest remedies typically work.

What’s the next event to watch for?

Watch the COFC docket for a ruling on TRAX’s complaint. It could come on a motion to dismiss or on the merits. Two other events matter: an Army or DoD public statement clarifying FAST TRACK’s governance, and another contractor filing a similar AI-hallucination claim at another agency.

BD classification: DEFEND Confidence: Moderate. It rests on one high-profile case plus a documented tool-adoption trend, FAST TRACK’s expansion beyond its origin unit. It is not yet a broad, multi-agency pattern. Evidence that would upgrade this: a COFC ruling ordering the Army to disclose FAST TRACK’s specific role, or ordering re-evaluation. Also, additional contractors filing similar AI-hallucination claims at other agencies. Evidence that would downgrade this: COFC dismisses the complaint on standing or merits without reaching the AI question. Also, Army documentation showing FAST TRACK’s use is tightly bounded and already disclosed as required.