Federal agencies’ growing use of artificial intelligence (AI) in procurement is opening a new front in bid protests. In Trax International Corp., No. 26-796 (Fed. Cl. Sept. 22, 2026), a federal appellate court rejected a challenge to a $449 million Army contract because Trax did not connect the Army’s use of AI to evaluation errors or show why reliance on AI undermined the award decision. The court noted that portions of the Army’s evaluation were nearly identical to an AI-generated assessment but held that “more is needed” to establish improper reliance. The ruling is narrow, but it signals that AI-generated material may become part of the protest record: the court required the Army to produce AI-generated evaluations, despite the government’s initial position that they were not part of the decision-making process.
The central question for future protests may be less whether an agency used AI and more whether the record shows meaningful human judgment. FAR 15.308 requires the source selection authority to exercise “independent judgment” and document the award decision. AI may help agencies triage or summarize lengthy proposals, but evaluators cannot simply adopt its output, particularly if it contains hallucinated references, erroneous conclusions, or findings inconsistent with the solicitation. A record that fails to explain how evaluators reached their ratings may give protesters a viable challenge even without proof that AI caused a specific error.
For contractors, the case points to practical questions for developing a protest: what AI-generated evaluations or summaries were created, who received them, and how were they checked against the proposals and solicitation? The court’s separate October 5, 2026, order seeking additional information about government declarations also underscores the importance of a complete and accurate record. Agencies that document how evaluators reviewed AI outputs and reached their own conclusions may be better positioned to defend an award; unsupported or unexplained AI-generated language in the evaluation may leave a meaningful opening for challenge.