The Army's acquisition supplement tells its contracting officials they "shall use" a bot called the Determination of Responsibility Assistant (DORA) "to assist them to determine prospective contractor responsibility or non-responsibility." That line in AFARS 5109.103 shows where AI in federal acquisition stands in 2026: the tools are in the workflow, sometimes by rule, and the decision still belongs to a named official. Below: where agencies report using AI by stage, which decisions the FAR keeps with a person, and what OMB's two AI memos ask of acquisition teams.
Where is AI used in federal acquisition today?
Mostly in the high-volume, rules-based work around each decision: screening documents, classifying data, searching files and drafting first versions. The clearest public record is the General Services Administration's (GSA) 2025 AI use case inventory, last updated September 10, 2026. The table adds two tools documented elsewhere, marked as such.
| Stage | Example tool (agency) | What it does, per the source | Status listed |
|---|---|---|---|
| Requirements | RAG AI for outcome-based contracting (GSA) | Tests whether an AI assistant can coach teams writing outcome-based requirements | Pre-deployment, experimental |
| Market research and pricing | LaborMatch IQ (GSA) | Uses retrieval-augmented generation for services pricing market research | Pre-deployment |
| Solicitation | Solicitation Review Tool (GSA) | Screens SAM.gov solicitations for Information and Communications Technology (ICT) notices that may lack required compliance language | Deployed |
| Evaluation | CALI (GSA) | Checks proposals against solicitation requirements, reps and certs and forms | Production as of May 2022 (ACT-IAC entry; not in the 2025 inventory) |
| Responsibility | DORA (Army) | Assists the responsibility determination | Required by AFARS 5109.103 |
| Contract file | ASSIST ECF Autoclassification and Summary (GSA) | Helps contracting officers sort the 20 to 50 documents in an electronic contract file | Pre-deployment |
| Spend analysis | Acquisition Analytics (GSA) | Classifies transactions in the government-wide category management taxonomy | Deployed |
The two deployed GSA tools check or classify against a fixed standard, which is where errors are easiest to catch. The generative tools that draft or coach are still in pre-deployment. A sample project in GSA's June 3, 2026 Elimination, Optimization and Automation (EOA) Playbook follows the same line: it proposes a robotic process automation bot to do the SAM.gov and FAPIIS research behind a responsibility determination, so contracting officers can "spend more of their time on the actual evaluation component."
Can AI evaluate proposals?
In the documented cases, AI checks proposals; it doesn't rate them. In the ACT-IAC entry for CALI, the tool compares a proposal against the solicitation's requirements, forms and representations, and the evaluator decides whether the offeror is compliant. Strengths, weaknesses, ratings and the award decision stay with the evaluation team and the source selection authority.
The harder question in practice is the data. Proposals carry contractor proprietary information, and evaluation records are source selection information. Before either goes into any tool, generative or not, confirm with your Chief Information Security Officer and counsel that the tool is authorized for that data, and write the tool's role into the source selection plan so evaluators know what they may rely on.
Which acquisition decisions do the rules keep with a person?
At minimum, the responsibility determination and the source selection decision. FAR 9.103(b), in the current FAR text, says "no purchase or award shall be made unless the contracting officer makes an affirmative determination of responsibility." DORA can gather and organize the record; the determination is still the contracting officer's.
For source selection, the Revolutionary FAR Overhaul's Part 15 model deviation text, at 15.205, says the decision "must represent the SSA's independent judgement," must be based on a comparative assessment against all the criteria in the request for proposals, and must document "the rationale for any business judgments and tradeoffs." Your agency may be operating under FAR overhaul class deviations for Parts 9 and 15, so check which version applies to your buy.
AI output enters the file as one more input, and the documented reasoning has to be the official's own. A decision memo that repeats a tool's summary without showing how the source selection authority (SSA) weighed it is hard to defend in a protest. How a tool's output can be used in a specific evaluation is for the contracting officer and agency counsel to decide, ideally before the evaluation plan is final.
Is AI used in acquisition "high-impact" under OMB M-25-21?
M-25-21 doesn't name procurement either way, so the answer depends on how a tool's output is used. Office of Management and Budget (OMB) Memorandum M-25-21, issued April 3, 2025, defines high-impact AI as AI "with an output that serves as a principal basis for decisions or actions" with "legal, material, binding, or significant effect" on, among other things, an entity's civil rights, access to programs or access to critical government resources or services. High-impact uses must meet minimum practices, including pre-deployment testing, an AI impact assessment, ongoing monitoring and human oversight.
Section 6 lists 15 purposes presumed to be high-impact, from critical infrastructure safety to federal employment decisions. Procurement is not among them, and the memo calls the list "not an exhaustive list." For uses on that list, an official who decides a use is not high-impact must document it in writing to the Chief AI Officer (CAIO). For other uses, the question is whether the tool meets the definition.
A tool that sorts contract files and a tool whose score becomes a principal basis for excluding an offer can land in different places. That call belongs to the agency, through its CAIO and AI governance board, which M-25-21 says must include procurement representatives when relevant. The memo gave agencies 365 days, to April 3, 2026, to document their minimum practices, so a new use that qualifies needs them in place before anyone relies on it.
What does OMB M-25-22 ask when an agency buys AI?
M-25-22, issued the same day, covers AI systems and services agencies acquire. It excludes common commercial products with embedded AI, such as a word processor, AI used incidentally by a contractor during performance, and AI acquired as part of a National Security System. It applies to contracts awarded from solicitations issued on or after September 30, 2025 (180 days after issuance), and to options to renew or extend exercised after that date.
The contract terms carry most of the weight. Section 3(e) asks agencies to ensure contracts "permanently prohibit the use of nonpublic inputted agency data and outputted results" to further train publicly or commercially available AI algorithms, absent explicit agency consent. Section 4 calls for terms that reduce vendor lock-in (knowledge transfer, data and model portability) and let the agency "regularly monitor and evaluate" performance and risk, and says data used for independent evaluation should not be accessible to the vendor.
GSA's own AI clause, GSAR 552.239-7001, sets similar data rules for large language model purchases; see our post on what the GSA AI clause requires.
What changes for vendors writing proposals?
Compliance checks are the part most likely to touch your proposal first, and they reward the basics. Mirror the solicitation's section numbering and terms, answer every "shall" where the instructions say to, and complete every required representation and form.
Then read for AI terms. Section 3(h) of M-25-22 warns about "the unsolicited use of AI systems by vendors" and asks agencies to determine when a solicitation should require disclosure of AI use in contract performance. If you deliver AI capability, expect data rights, training and monitoring terms. If you plan to use AI in performance and the solicitation is silent, ask the contracting officer during the question period rather than assuming.
How acquisition teams can start
Start where the volume is and the standard is fixed: clause and compliance checks, contract file organization, spend classification, first drafts of routine documents. Those are the uses GSA lists as deployed or closest to it, and their errors show up quickly in review.
For each use, write down who reviews the output, what they check it against and where that review is recorded in the contract file. Ask your CAIO's office whether the use needs a high-impact review, and keep the determinations with the officials FAR 9.103 and 15.205 name. If cycle time is the goal, measure it from a baseline; our post on procurement administrative lead time explains what the standard PALT measure does and doesn't count.
For help planning that work, talk to our analytics and AI team.
The Bottom Line
AI is already doing the sorting, screening and first drafts in federal acquisition, and the rules still put each binding decision on a named official who has to explain it. Put AI where the volume is, write down who checks its output before a decision relies on it, and ask your Chief AI Officer's office early whether a use needs high-impact review.
FAQ
Can AI award a federal contract?
Not under the FAR as written. FAR 9.103(b) says no purchase or award shall be made unless the contracting officer makes an affirmative determination of responsibility, and the source selection decision must represent the source selection authority's independent judgment. AI can prepare research and analysis that those officials review.
Do agencies use AI to evaluate proposals?
Some have used it for compliance checks. A May 2022 ACT-IAC use case entry describes GSA's Contract Acquisition Lifecycle Intelligence (CALI) tool, which checks proposals against solicitation requirements and leaves the evaluator to decide whether the offeror is compliant. Ratings and the award decision remain with the evaluation team and the source selection authority.
What does OMB M-25-22 require?
M-25-22, issued April 3, 2025, sets rules for agencies buying AI systems and services. It applies to contracts awarded from solicitations issued on or after September 30, 2025, and to options exercised after that date. It calls for contract terms on data rights, a permanent bar on using nonpublic agency inputs and outputs to further train publicly or commercially available AI without consent, protection against vendor lock-in and the agency's ability to monitor performance.
Is AI used in procurement considered high-impact AI?
M-25-21 does not list procurement among the uses presumed to be high-impact, but it says the list is not exhaustive. High-impact AI is AI whose output serves as a principal basis for decisions with legal, material, binding or significant effect on listed rights, resources and services. Whether a given acquisition tool meets that test is for the agency and its Chief AI Officer to determine.
Do contractors have to disclose their own AI use on a federal contract?
Check the solicitation and contract. M-25-22 excludes AI a contractor uses incidentally during performance from its scope, but asks agencies to determine when a solicitation should require disclosure of AI use in contract performance. The contract's terms, as read by the contracting officer, decide what you must disclose.
Sources
- OMB Memorandum M-25-22, Driving Efficient Acquisition of Artificial Intelligence in Government (April 3, 2025)
- OMB Memorandum M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust (April 3, 2025)
- GSA, 2025 GSA AI Use Cases (inventory, updated September 10, 2026)
- AFARS 5109.103, Policy (Determination of Responsibility Assistant)
- FAR 9.103, Policy (responsible prospective contractors, FAC 2026-01)
- Revolutionary FAR Overhaul, Part 15 model deviation text (15.205, source selection decision)
- ACT-IAC, Contract Acquisition Lifecycle Intelligence (CALI) tool use case (May 2022)
- GSA, Federal EOA Playbook (June 3, 2026)
How we researched this. Every number, date and requirement here is checked against the sources listed above, primary documents first, and the post is fact-checked again before publication. When a rule, date or status changes, we update the post and log the change below. Read our editorial standards or send a correction.





