GSA is openly rewriting the rules
In a February 2026 interview, Jeffrey Koses, GSA's senior procurement executive, confirmed something most of the federal contracting community already suspected: AI is forcing GSA to rewrite solicitation design, evaluation methodology, and contract terms. GSA has launched nearly two dozen OneGov agreements covering AI tools from Anthropic, OpenAI, Google, Microsoft, Meta, xAI, and Perplexity. And Koses publicly warned that AI is creating new acquisition risks — including a measurable rise in protests and proposal filings that contain AI-generated, fabricated citations.
That last point is worth dwelling on. When the senior procurement executive at GSA volunteers that the agency is seeing fabricated case citations in protests, you are watching a real shift in how acquisitions are run. Here is what is actually changing and what contractors should be doing about it.
Three shifts in federal acquisition
1. AI-specific terms and conditions are coming
GSA is standardizing AI-related terms across OneGov agreements and is preparing to push them into the broader MAS solicitation. Expect new language addressing:
- Model provenance and training-data disclosure. Where did the model come from, and what was it trained on?
- Bias, hallucination, and accuracy representations. What testing has been done? What guardrails are in place?
- Customer data handling. Will the model train on government data? Where is the data stored? Who has access?
- Government use rights and IP. Standard FAR/DFARS data rights frameworks need adaptation for foundation models, embeddings, and fine-tunes.
If you sell anything with an AI component on a federal contract, draft your model and data disclosures now. When the new clauses land, contractors who already have answers will breeze through. Contractors who do not will scramble at modification time.
2. Solicitation design is shifting
Koses described AI as "lowering proposal costs and expanding the pool of bidders." That is a polite way of saying AI is making it cheap to generate proposal volume. From the source selection authority's perspective, that creates two problems: more proposals to evaluate, and proposals that look superficially compliant but lack underlying substance.
Expected solicitation design changes:
- Tighter compliance gates. Section L instructions will get more prescriptive — not because the government wants more paperwork, but because tighter instructions filter out poorly-targeted AI-generated submissions faster.
- More narrowly-defined past performance criteria. Same logic — tighter relevance standards filter out generic submissions.
- Page limits and word counts will be enforced more strictly. AI makes it trivial to expand text. Page limits put a cap on the volume problem.
- Live demonstrations and oral proposals will expand, especially for IT and AI-related buys, because they are harder to game with AI-generated content.
If you are bidding under solicitation language that looks more prescriptive than what you remember from two years ago, that is not random. It is a deliberate response to AI-driven proposal volume.
3. Protests are getting weirder
Koses' comment about AI-generated protests with fabricated citations is the most concrete data point in his interview. The federal courts have already sanctioned attorneys in unrelated litigation for filing AI-generated briefs with fabricated case citations; the same problem is now showing up at GAO and the Court of Federal Claims for protests.
For contractors, the implication has two sides:
- Defending against weak AI-generated protests. If a competitor files a protest of your award and you can identify fabricated citations or non-existent precedent in the protest, that is grounds for swift dismissal and, in some cases, sanctions. Your protest counsel should be checking citations.
- Filing your own protests responsibly. The temptation to use AI to draft a protest filing on a tight deadline is real. Resist it without aggressive citation-checking. Sanctions for fabricated citations land on the firm and the named attorney — not the AI.
What contractors should do
Five concrete moves for the next two quarters:
- Audit your AI footprint on every active contract. For each contract, identify whether AI is part of the delivered product or service, even tangentially. If it is, document the model, data flows, and customer-data handling now.
- Prepare your standard AI disclosures. A one-page document covering model provenance, training data, bias testing, accuracy benchmarks, and data handling policies. You will need it for the next solicitation that asks.
- Tighten your proposal compliance discipline. Treat every Section L requirement as a hard gate. Build a compliance matrix. Have someone outside the proposal team review against the matrix before submission. AI-driven solicitation tightening is going to punish loose compliance.
- Build a protest-citation checklist. If you protest, every cited case, regulation, and decision needs verification. If you are protested against, run the same check against the protest you receive — fabricated citations are now a real defense.
- Read OneGov AI agreements for your sector. GSA has published terms for some OneGov agreements. Read them. The terms in those agreements are the template for what GSA will eventually push to MAS contractors.
When I was a CO, technology shifts always favored prepared contractors
I have seen multiple technology cycles inside acquisition shops — the move from paper to email, the move from sealed bidding to FAR Part 15 negotiations on more buys, the move from GSA Advantage to FCP. Every cycle produced a small group of contractors who read the early signals and adapted, and a larger group who waited for the rules to settle and discovered they had been outmaneuvered.
AI is the most disruptive shift I have seen for federal acquisition in 20 years. The contractors who get this right will win disproportionate share through 2027 and 2028.
A note on AI in your own proposal process
Two practical guardrails:
- Use AI to draft, not to certify. AI can speed up your first draft, your past performance summaries, and your compliance matrix. It cannot certify your representations, your past performance accuracy, or your pricing. Those are signed by a human officer of your company under penalty of False Claims Act exposure. Treat them accordingly.
- Document your AI use internally. If a CO asks at modification time whether your firm used AI in proposal preparation — and that question is coming — have an answer that distinguishes drafting from certification.
Bottom line
AI is no longer a future federal acquisition issue. GSA is openly rewriting solicitation strategy, OneGov AI agreements are setting the template for new clauses, and the protest forum is already seeing AI-generated filings with fabricated citations. The contractors who will win the next two years are the ones who treat AI as both a tool and a compliance risk — using it to accelerate proposal work while building the disclosure, citation discipline, and contract administration practices that the new federal acquisition environment will require. Audit your AI footprint, prepare your disclosures, and tighten your compliance discipline before the next solicitation forces you to.