The Government Accountability Office’s (GAO) bid protests and appropriation section is currently furloughed pending a resolution of the partisan bickering that has shutdown federal government. But on the other side of the agency pause button is a likely pressure cooker of protests ready to file day one when the GAO door reopens for business. Agency level protests can only dam the rising tide of complaints for so long (if anyone is around to access those). With the incoming flood of protests likely coming, many self-prepared, it is important to note some recent decisions from the GAO regarding the reliance on artificial intelligence. That is, using Chat GPT or some other AI software, to handle the vital protest drafting. These tools are handy and can be an incredibly helpful and cost saving tool, but it is important to understand and avoid the pitfalls that can lead to an immediate protest dismissal – or worse.
The GAO has fired a clear message across the bow to contractors and counsel alike: artificial intelligence cannot replace human verification. In a series of recent bid protest decisions, culminating in the brutal Oready, LLC (B-423649 et al., Sept. 25 2025), GAO has dismissed filings or issued stern warnings where parties submitted briefs containing ghost or misrepresented citations that bore “the hallmarks of AI-generated content.” The result: GAO’s first published dismissal for abuse of process tied to AI misuse to support a growing line of cases establishing a “trust-but-verify” rule for AI in federal procurement litigation.
The Oready Decision: A Bright-Line Moment
In Oready, LLC, the protester challenged Bureau of Indian Education solicitations requiring signed commitment letters, on-site service obligations, and other key-personnel restrictions. GAO never reached the merits.
Instead, GAO found that Oready’s pro se filings contained:
- Entirely fabricated GAO decisions (invented B-numbers and captions);
- Mismatched citations, pairing legitimate case names with wrong numbers; and
- Misquoted holdings from real decisions.
GAO concluded these errors reflected a “reckless disregard for accuracy,” not simply innocent mistakes, and subsequently dismissed the protests as an abuse of GAO’s process. The decision stressed that while GAO was not sanctioning the mere use of AI, it would not tolerate filings that appeared to rely on AI without human verification:
“Our decision here is not based on the use of AI as a method of research; it is based on the protester’s repeated reliance on non-existent citations or decisions without verifying their correctness and applicability to the arguments being presented.”
GAO reaffirmed its inherent authority to dismiss protests or impose sanctions when parties undermine the integrity of the forum. The lesson, which is a constant in today’s legal news, is that even AI needs an editor. It has always been imperative to check cites and this process does not simply fade into history with the use of AI, which remains to have a fondness for Timothy Leary’s favorite pastime.
The Pattern: GAO’s Growing Line of AI-Related Warnings
Oready is not an isolated event, but the crest of a problem GAO has been dealing with consistently as of late. Since early 2025, GAO has flagged multiple protests for citation irregularities linked to unverified AI drafting:
| Case | Date | AI / Citation Issue | Outcome |
|---|---|---|---|
| Raven Investigations & Security Consulting, LLC (B-423447) | May 7, 2025 | Filed with “citations to purported GAO decisions that could not be found.” | Protest dismissed; GAO issued its first apparent AI-related warning about sanctions. |
| Wright Brothers Aero, Inc. (B-423326.2) | July 7, 2025 | Protester admitted an erroneous citation was introduced via AI software. | Protest dismissed as untimely; GAO formally warned in a footnote that continued misuse may lead to sanctions. |
| BioneX, LLC (B-423630) | July 25, 2025 | Citations had “hallmarks of cases generated by AI.” | Dismissed for failure to state a claim; GAO reiterated sanctions warning. |
| IBS Government Services, Inc. (B-423583) | Aug 20, 2025 | Filings included fabricated or misquoted GAO cases. | Denied on merits, but GAO repeated its Raven warning verbatim. |
| Oready, LLC (B-423649 et al.) | Sept 25, 2025 | Fabricated decisions, misnumbered citations, false holdings. | Dismissed for abuse of process. |
GAO’s language has grown progressively sharper. Early decisions used “cautionary” language but by Oready, just a few months down the road after a patten because clear, the tone shifted to enforcement. The agency now routinely notes that protest filings with the “hallmarks of unverified AI generation” may be dismissed or sanctioned outright.
AI and the Accountability Trend Inside GAO
These protest decisions are part of GAO’s broader focus on AI governance across the federal landscape. In recent oversight report, including GAO-24-105980 and GAO-25-107653, GAO has evaluated how agencies manage and monitor AI tools. It also released an AI Accountability Framework (GAO-21-519SP) that sets four pillars for responsible AI: governance, data quality, performance, and ongoing monitoring.
Ironically, GAO now applies these same principles—transparency, traceability, and verification—to its own protest docket.
GAO has even begun experimenting with large-language models internally to summarize records and analyze legislative text. But its message to outside litigants remains consistent: AI is acceptable only when verified by humans.
Why It Matters: The Integrity of the Forum
GAO’s bid protest process depends on the accuracy and reliability of the record. Filing a protest with false citations—whether produced by AI or human error—undermines that trust.
The agency’s decisions make three principles clear:
- AI is not the enemy—laziness is. The problem is not the use of generative tools, but failure to fact-check and edit output.
- The standard applies to everyone. GAO explicitly states that being pro se or unsophisticated does not excuse citation errors.
- Sanctions are on the table. GAO will use its inherent authority to preserve the “inexpensive and expeditious” protest process Congress envisioned.
How to Stay Safe: AI Verification Protocol
Whether you are in-house counsel, a government contractor, or outside protest counsel, build a verification gate before filing anything touched by AI.
Four-Step Verification Checklist
| Step | Action | Purpose |
|---|---|---|
| 1. Primary-Source Pull | Download every decision directly from gao.gov or govinfo.gov. | Avoid secondhand errors from summaries or AI tools. |
| 2. Verify & Log | Record B-number, case name, date, and proposition. | Creates a defensible record of diligence. |
| 3. Human Cite-Check | Have a second person verify each authority before filing. | Adds a manual quality-control layer. |
| 4. Correct Quickly | If you spot a mistake post-filing, notify GAO and correct it immediately. | Demonstrates candor and mitigates risk of sanctions. |
The Path Ahead
AI tools will remain part of legal drafting and research. The tools provide substantial time and cost savings in many instances. GAO itself uses them as do other agencies in the federal government. But Oready and its companion cases illustrate a new professional standard: AI-assisted work must meet the same ethical and factual accuracy requirements as human work and a simple cite check is usually the easiest path to creating successful protest. Find the correct cite, rely on the correct language and propositions stated in the matter.
As GAO continues to oversee federal AI adoption, expect the same accountability principles to govern procurement, oversight, and litigation alike. In other words, AI may complete the proverbial first draft of your protest, but you must be the editor, verifier, and guarantor.
