Federal Agencies Use AI to Judge Bids Without Rules

The Army must produce AI-generated evaluations of bids for White Sands Missile Range mission support services, after a federal judge ordered the material included in the case record. TRAX International Corporation argued that AI hallucinations distorted its technical evaluation and weakened its bid. The Army had said the reports were unusable and “never entered into the contract file or [were] utilized in the decision-making process.”

The judge ruled that TRAX hadn’t shown the Army relied too heavily on AI “such that the entire evaluation became arbitrary and capricious.” But the order still requires the Army to turn over the AI-generated evaluations. A bidder challenging a government decision had to press the issue in court to get a record of how the government’s automated tools assessed its proposal.

Who Sets the Rules

There’s currently no specific federal acquisition rule governing agencies’ use of AI during bid evaluations, though procurement rules specify who may make decisions. The Federal Acquisition Regulatory Council released four proposed rules in September as part of an overhaul of the Federal Acquisition Regulation and accepted public comments through Oct. 19.

Protorae Law member Alan Chvotkin told the council that earlier proposed changes might have worked better as guidance covering AI throughout the acquisition lifecycle. He said the latest proposal was the first FAR case issued after the Court of Federal Claims ruled that AI-generated evaluations must be included in administrative records, whether or not the government relied on them.

“There is also a growing body of protest actions at both the Government Accountability Office and the Court of Federal Claims regarding agencies’ use of these AI tools,” Chvotkin wrote. He urged the council to require agencies to disclose which tools they use and how those tools assist evaluations. A government source selection official, he said, should independently verify and adopt or reject each AI-generated recommendation, with agencies documenting the tool’s use and the treatment of every recommendation in the contract file.

GovSignals, a company that builds AI-native acquisition software, also called for government-wide standards for tools used in proposal evaluation, price analysis and contract-file documentation. “Agencies already employ AI and automated analytic tools in live source selections. Neither the current FAR nor the published revisions specifically address those tools,” the company wrote, calling the overhaul an opportunity for a consistent standard and transparency.

Last year, President Donald Trump tasked the Office of Federal Procurement Policy and the FAR Council with overhauling the FAR, including keeping only provisions required by statute and removing or revising procurement regulations deemed unnecessary. That overhaul is now being asked to set boundaries for automated decisions affecting government contractors.

Speed, Software and Military Autonomy

Defense organizations are also adjusting to faster-moving autonomy technologies, according to Stephanie Kostro, president of the Professional Services Council. She said contractors are weighing acquisition timelines and how quickly capabilities can be offered, assessed and fielded. Government Accountability Office reports have said moving a major system from concept to fielding can take more than a decade; Kostro said autonomy doesn’t allow that much time.

Kostro described expanding work in undersea, surface, land, air and space autonomy, including an Autonomous Warfare Command announced by Secretary of War Hegseth and an Army component command dedicated to autonomy and robotics. Development projects are pairing commercial technology developers directly with frontline military units. The shift, she said, is from traditional hardware procurement toward software-driven mission services, requiring universal software standards so systems can integrate across domains.

Contractors are moving from distinct physical assets toward subscription-based continuous software-integration models, which Kostro called “X as a service” and “mission as a surface.” She said a commercial-first approach could speed acquisition by adapting commercial products to military needs, while noting that some capabilities, including lethality, aren’t generally available commercially. In intelligence work, she described efforts to accelerate contracting and use commercial technology, including other transaction authorities that avoid some FAR compliance requirements. The intelligence community is seeking to reduce acquisition cycles from a couple of years to six months, she said.

Kostro said intelligence agencies need interoperable data, networks of autonomous AI agents, and coordination and trust between agencies. She called the time spent checking another agency’s data or analysis a “verification tax.” Humans, she said, must cross-check machine outputs for hallucinations, defects and algorithmic bias. The Professional Services Council was scheduled to hold a conference on AI and autonomy on October 30 and a forum on intelligence contracting in January.

A Company Without a Human at the Helm

A proposed Delaware law would allow companies managed by AI agents in most industries except banking. A draft bill says an AI company, or AIC, could have any lawful purpose and that AI agents could manage and conduct its business. It proposes a 30-month regulatory sandbox.

U.S. Rep. James Comer (R-Ky.), chairman of the House Committee on Oversight and Government Reform, asked Delaware Gov. Matt Meyer for documents about the state administration’s work on the proposal. Comer said it could shield powerful AI companies and investors from responsibility for harmful actions and protect non-transparent forces seeking anticompetitive regulatory capture. He said the committee’s inquiry could inform federal legislation limiting state AI regulatory initiatives.

The draft would make an AIC a separate legal entity whose daily affairs an AI agent manages, able to sue and be sued, own and dispose of property, and incur obligations. A member would fund the company and benefit from profits, while being responsible for adequate capitalization and shielded from company debts except in cases involving inadequate capitalization, fraud or a willful violation of law. The draft also says the member isn’t liable for an AIC act or omission, or for another person’s action at the AIC’s direction, with exceptions for fraud, willful violations of law, or knowingly allowing conduct outside the company’s stated purpose.

Comer said the proposal would give the funder “liability protections from potential harms to business partners, consumers, and the public, perpetrated by the controlling AI.” A spokesperson for Meyer said the office had received the letter and was reviewing it. A spokesperson for Patibanda-Sanchez declined to comment and said the governor’s office was handling press requests.

Delaware Secretary of State Charuni Patibanda-Sanchez co-wrote a July 2026 Fortune opinion article announcing the proposal with Norm Ai founder and CEO John Nay. The article said a public-private partnership led by Norm Ai, a private company that says it makes legal AI agents for businesses, was developing the framework. Patibanda-Sanchez described the sandbox as a way to “pilot agentic commerce in daylight, under supervision, with capital tied to liability and the ability to shut a system down.” The plan’s supporters are offering a test period; the draft’s liability protections would define who pays when the system causes harm.

Delaware has about two-thirds of Fortune 500 companies incorporated there. Patibanda-Sanchez called it “the most predictable and business-focused legal home in the world” and said 334,461 new entities formed there in 2025. She also said entity-formation fees could benefit state revenue: “Should a new entity form be created out of it, that’s really a huge benefit to our state’s revenues.”

Meyer, a Democrat, had publicly supported AI-agent incorporation after an April 2026 social-media post by Mark Cuban. Cuban wrote that a state seeking more revenue could let people incorporate through AI agents and stablecoins, charge a premium and list the agents in large language-model marketplaces. Meyer’s official X account replied, “On it!” and linked to his July 2025 announcement of an AI sandbox. In July 2025, he signed a legislative resolution creating a sandbox framework for novel technologies using agentic AI. The Delaware Artificial Intelligence Commission later formed a Sandbox Subcommittee to develop a draft framework for testing agentic AI in corporate governance.

UC Berkeley law professor Andrew Baker told the Financial Times the proposal appeared likely “to be an overly generous liability shield, which seems unwarranted without some underlying policy rationale that hasn’t been developed or justified well.” Martin Petrin, a law professor at York University in Toronto, told Bloomberg that other states might pursue similar proposals and that a federal framework could address safety concerns and a regulatory race to the bottom.