FAR Council petitioned over automated evaluation errors in US Army contract

A bid protest over US Army technical evaluations has prompted procurement attorneys and industry groups to demand formal federal regulations for automated reasoning systems.

The dispute, unsealed in federal court, revealed contracting officers relied on automated tools that introduced factual errors during proposal evaluations. The system mischaracterised vendor submissions, attributing compliance deficiencies to proposals that did not exist in the source files. The findings led to the disqualification of a bidding contractor before manual reviews caught the discrepancies.

Court disclosures have galvanised acquisition lawyers and trade associations to petition the Federal Acquisition Regulatory Council. The coalition argues that existing procurement guidance fails to address generative model hallucinations and non-deterministic scoring mechanisms. They are seeking binding revisions to the Federal Acquisition Regulation that would standardise how agencies deploy automated evaluation software.

Under current rules, agency acquisition teams may experiment with commercial artificial intelligence tools to accelerate proposal reviews. Federal directives encourage automation to clear administrative backlogs. However, the policies establish few enforceable thresholds for model validation, data provenance, or auditability during competitive source selections.

Legal specialists warned the FAR Council that unverified machine outputs threaten the integrity of the procurement system. When algorithms generate synthetic rationales, contracting officers risk issuing arbitrary and capricious award decisions that cannot withstand judicial scrutiny. Petitioners argue that agencies must log all model prompts, maintain strict configuration controls, and ensure human evaluators corroborate every factual assertion against raw vendor filings.

The proposed regulatory changes would require contracting officers to disclose when artificial intelligence assists in drafting technical consensus reports. They would also mandate standard testing protocols to detect hallucinated text before agencies eliminate offerors from the competitive range.

The FAR Council has not yet issued a formal response to the petition. With defence and civilian agencies expanding their use of automated workflows, procurement courts will face mounting pressure to define the legal boundaries of algorithmic decision-making. Can the federal regulatory apparatus establish enforceable verification baselines before widespread agency adoption renders commercial bids vulnerable to automated confabulation?