OMB memorandum on acquisition practices for artificial-intelligence systems and services in government.
Federal AI Procurement
Accountability file
A public-record case file for federal AI adoption, acquisition, governance, risk-management, and procurement-integrity records. It keeps policy memoranda, acquisition rules, standards, spending records, vendor references, and coverage leads separate until a public record supports the exact claim.
- Updated
- Jul 18, 2026
- Source Anchors
- 14 verified · 188 leads
- Timeline
- 12 events
- Roles
- 6 public-record
- Fingerprint
409C E95D 89B4
How the public record connects
This shows the case file's most significant public-record connections only — see the Role Web and Timeline tabs for the full set. A line means a record connects two points; it is not a finding about conduct.
Publishing limits
These are the standards every anchor on this file must clear.
- Minimum source tier
- official record
- oversight report
- spending record
- recipient integrity record
- legislative record
- Living people
- For named officials, employees, or vendors, publish only source-backed public roles and exact record relationships; do not infer intent, influence, or blame.
- Victim protection
- Do not publish private bidder data, nonpublic proposal material, source-selection details, or personal contact information.
- Social leads
- Social posts and coverage can start a procurement lead, but award-level copy requires official procurement, spending, agency, court, protest, or enforcement records.
No guilt by proximity
A vendor, model, agency, or policy record appearing in the same procurement lane is context only and does not prove an award, conflict, deployment, or failure.
People, institutions, and the records between them
Every public-record role in this file, grouped by kind. Open a role to read its dossier — its public-record status, what it is, and the exact records that name it. A shared record naming two roles is not a finding about conduct beyond that record.
Every public-record role
Pick a role to see its public-record status, what it is, and the exact records that name it. A shared record is not a finding about conduct.
OMB memorandum on accelerating federal AI use through innovation, governance, and public trust, with implementation instructions for agency AI use.
GAO report titled Artificial Intelligence Acquisitions: Agencies Should Collect and Apply Lessons Learned to Improve Future Procurements, published and publicly released April 13, 2026, carrying four recommendations for executive action.
Federal Acquisition Regulation part covering improper business practices and personal conflicts of interest.
Federal Acquisition Regulation part covering acquisition policies and procedures for information technology.
GAO report titled ARTIFICIAL INTELLIGENCE: Generative AI Use and Management at Federal Agencies, published and publicly released July 29, 2025, covering generative-AI use cases, agency inventories, challenges, and government-wide policy context.
NIST AI RMF record for voluntary risk-management guidance and trustworthiness considerations.
Earlier OMB memorandum establishing acquisition-related practices for AI systems and services.
Timeline
- public document repository / official#NIST AI RMF baseline enters the source laneNIST's AI Risk Management Framework becomes the standards-context anchor Hugin can use when reviewing public AI procurement and deployment records.Standards context does not establish whether a specific agency or vendor met the framework.
- public document repository / official#NIST generative-AI profile enters acquisition reviewNIST's generative-AI profile gives Hugin an official risk-management crosswalk for AI acquisition and deployment review without naming a specific agency system.Risk-management profile context does not establish whether a specific agency or vendor met the profile.
- records release review / official#OMB AI acquisition guidance enters reviewOMB M-24-18 gives Hugin a prior acquisition-policy record for AI systems and services before later 2025 guidance changed the posture.A policy memo is not evidence that a named AI contract was awarded or mismanaged.
- records release review / official#OMB updates AI use and acquisition guidanceOMB M-25-21 and M-25-22 create the current federal AI use and acquisition guidance lane for Hugin's procurement watch.Updated guidance does not identify a specific vendor, award, or deployment without separate records.
- records release review / official#White House AI Action Plan adds adoption contextThe White House AI Action Plan adds federal adoption and interagency-coordination context, while Hugin keeps recommendations separate from completed procurement events.Recommendations and policy direction are not award-level spending records.
- records release review / official#Executive Order 14319 sets federal LLM procurement termsOn the same day as the AI Action Plan, Executive Order 14319 was signed, directing agency heads to procure only large language models developed in accordance with two principles the order names the Unbiased AI Principles, giving OMB 120 days to issue implementing guidance, and directing agencies to include contract terms charging decommissioning costs to a vendor terminated for noncompliance. It was printed in the Federal Register on July 28, 2025 at 90 FR 35389.The order is a procurement mandate on agencies; it names no model, developer, or contract, and this file records no finding that any company or agency has complied or failed to comply with it.
- oversight finding / official#GAO generative-AI use oversight addedGAO's generative-AI use and management report adds oversight context for agency inventories and policy implementation, while Hugin keeps inventory findings separate from award-level procurement claims.Agency inventory oversight does not prove a particular AI acquisition, deployment, or vendor issue.
- records release review / high#M-25-22 attaches to new AI solicitationsBy its own terms, OMB M-25-22 applies to any contract awarded pursuant to a solicitation issued on or after the date that is 180 days after its April 3, 2025 issuance. Counted from that issuance date, the trigger falls on September 30, 2025. The date is derived from the memorandum's text rather than printed in it, and this desk has checked no individual solicitation against it.A policy applicability date is not evidence that any particular solicitation, award, or agency was covered by the memorandum, complied with it, or failed to.
Public sources ready to ingest
Import sources create source-run packets, document inventories, and review tokens. They do not publish accusations or graph relationships until the case-file evidence rules are satisfied.
Source anchors
OMB memorandum on accelerating federal AI use through innovation, governance, and public trust, with implementation instructions for agency AI use.
Use as a federal AI-use policy anchor; do not convert broad agency guidance into contract-specific deployment claims.official recordOMB M-25-22 AI acquisition memorandumOffice of Management and BudgetOMB memorandum on acquisition practices for artificial-intelligence systems and services in government.
Use for acquisition-policy posture; join to solicitations, awards, and agency records before naming procurement outcomes.official recordOMB M-24-18 AI acquisition memorandumOffice of Management and BudgetEarlier OMB memorandum establishing acquisition-related practices for AI systems and services.
Use as a historical acquisition baseline and compare against later OMB guidance before citing current requirements.official recordAmerica's AI Action PlanThe White HouseWhite House AI Action Plan record with federal AI adoption, infrastructure, and policy recommendations.
Use for administration policy posture; do not treat recommendations as completed procurement or deployment events.official recordExecutive Order 14319, Preventing Woke AI in the Federal Government (90 FR 35389)The White House / U.S. Government Publishing OfficeThe executive order directing that federal procurement of large language models take account of stated truthfulness and ideological-neutrality criteria, issued the same day as the AI Action Plan and building on EO 13960.
An executive order sets policy for the executive branch and directs how agencies buy. It is not a finding about any vendor or model, and the criteria it announces are contested policy terms rather than measured properties of any system.official recordNIST AI Risk Management FrameworkNational Institute of Standards and TechnologyNIST AI RMF record for voluntary risk-management guidance and trustworthiness considerations.
Use as standards context; do not treat voluntary framework language as proof that an agency or vendor passed a review.legislative recordFAR Part 39 information-technology acquisitionAcquisition.govFederal Acquisition Regulation part covering acquisition policies and procedures for information technology.
Use for acquisition-rule context; pair with the specific solicitation or contract clause before applying it to an award.legislative recordFAR Part 3 improper business practicesAcquisition.govFederal Acquisition Regulation part covering improper business practices and personal conflicts of interest.
Use as procurement-integrity context only; do not infer a conflict without a specific public record.official recordNIST AI RMF Generative AI ProfileNational Institute of Standards and TechnologyNIST profile extending the AI RMF for generative-AI risks, including risks and actions relevant to federal AI acquisition and deployment review.
Use as generative-AI risk-management context; do not treat the profile as an agency- or vendor-specific compliance finding.baselineWhat the record currently supports
Claim rows are derived from timeline events and source anchors. They are evidence-status labels, not accusations or final truth claims.
What should be released next
Release requests are lawful public-record asks. They identify source gaps, review gates, and privacy constraints; they are not evidence that a specific person committed wrongdoing.
What stays unresolved
Deliberate gaps in the public record — open until the underlying records are released or reviewed, not questions we have skipped.
- Filed
- Hugin · public social evidence reports
- License
- Text and data are shared under the terms noted at hugin.studio/support · attribution appreciated.
- Fingerprint
409C E95D 89B4
For researchers & contributorsSource desk, research joins & machine-readable exportsEverything Hugin uses to grow and audit this file — connection detail, review posture, scout queue, and every export.
Every machine-readable slice of this file
How the public record is joined
Connections are navigation aids. They show where a record, event, or role is sourced; they do not convert association, correspondence, testimony requests, or document appearance into misconduct findings.
Scout queue before promotion
Federal AI procurement desk for OMB guidance, NIST risk-management context, FAR acquisition rules, spending records, recipient integrity checks, and vendor identity joins.
Source leads and automated probes are review inventory. They become verified anchors only after Hugin can attach the public record, legal posture, and no-allegation guardrails.
Public-record joins are wired into exports
These are not loose TODOs anymore. Each lane below is computed from the case file's source tiers, then joined to matching sources, timeline events, role nodes, and release-roadmap requests.
Track OMB, White House, NIST, and acquisition-rule changes before saying current federal AI policy changed.
Compare AI acquisition records with NIST RMF context and agency-specific implementation records before publishing risk-management language.
Join USAspending award rows, SAM entity records, agency solicitations, and vendor identifiers before naming any contract-specific claim.
Hold conflict, protest, exclusion, or enforcement leads until a specific filing, award protest, disclosure, or enforcement record is attached.
Help grow the source queue
Hugin accepts public records, not accusations. A submitted URL lands in a review lane, gets source-validated, and only becomes a case anchor after the evidence rules and privacy gates are satisfied.
What this case can prove right now
Fact-check status is a review posture, not a verdict. Source-scout rows are leads until a reviewer joins them to publishable anchors; claim-ledger rows are timeline-derived and still require source validation plus no-allegation guardrails before promotion.
Queue a public record
Paste an official or public-record URL. Hugin classifies the lane first, keeps lead-only material quarantined, and requires source validation before anything reaches the case file.
Queue a public source for review.Latest submitted public URLs
Sanitized queue view: Hugin shows the public URL, lane, status, and why a source has not been promoted. Submitter identifiers are not shown.
Submit sources, not claims
A source is a public URL someone else can inspect. A claim is an interpretation. Hugin queues sources first and keeps claims out until a reviewed public record supports the exact wording.
Official agency pages, court dockets, congressional releases, civil records, state updates, and public repositories with source URLs.
Publisher, date, record family, and why the URL belongs in this file.
Source URLs first. Context is welcome, but claims do not publish from submissions alone.
Private tips, survivor identifiers, sealed or leaked material, graphic details, and private contact data.
Screenshots or social posts without a public source URL.
Misconduct labels for living people unless the exact cited public record supports that exact statement.
Case change ledger
Audit feed for source anchors, queue decisions, privacy-review jobs, and hashes. It explains the process without turning leads into claims.
FAR Part 3 improper business practices
- Status
- needs human privacy review
- Evidence state
- metadata only
- Lane
- privacy triage
Extract document index and source metadata first.
Next: Metadata extraction includes publisher, URL, document family, and date.2645027897adFAR Part 39 information-technology acquisition
- Status
- needs human privacy review
- Evidence state
- metadata only
- Lane
- privacy triage
Extract document index and source metadata first.
Next: Metadata extraction includes publisher, URL, document family, and date.0b6b5ce8ffbcCase file hash snapshot
- Status
- hash changed when case bundle changes
- Evidence state
- metadata only
Mirrors can compare this hash to verify the same case file bundle.
Next: Compare the short hash in the UI with this machine-readable ledger.409ce95d89b4Document manifest hash snapshot
- Status
- hash changed when document handling changes
- Evidence state
- metadata only
This binds document mirroring and privacy-triage rules.
Next: Compare the short hash in the UI with this machine-readable ledger.90fbb28218a3Document review queue hash snapshot
- Status
- hash changed when document jobs change
- Evidence state
- metadata only
This binds privacy-review and metadata extraction jobs.
Next: Compare the short hash in the UI with this machine-readable ledger.bf0f01c65570Source validation hash snapshot
- Status
- hash changed when queue or validation rules change
- Evidence state
- metadata only
This binds the visible queue decisions to a machine-readable report.
Next: Compare the short hash in the UI with this machine-readable ledger.2e9a6e107d1eSubmission queue hash snapshot
- Status
- hash changed when public queue rows change
- Evidence state
- metadata only
This changes when sanitized submission rows change.
Next: Compare the short hash in the UI with this machine-readable ledger.c249ae276610FAR Part 39 information-technology acquisition
- Status
- source anchor in case file
- Evidence state
- case anchor
- Lane
- legislative record
Use for acquisition-rule context; pair with the specific solicitation or contract clause before applying it to an award.
Next: Keep source-linked timeline and role claims limited to the extracted facts.b6381819b7edGAO artificial intelligence acquisitions report
- Status
- source anchor in case file
- Evidence state
- case anchor
- Lane
- oversight report
An oversight recommendation report, not an adjudication: GAO recommends policy changes at named agencies and makes no finding of misconduct against any agency, official, or vendor.
Next: Keep source-linked timeline and role claims limited to the extracted facts.f9de11fae20bGAO generative AI use and management report
- Status
- source anchor in case file
- Evidence state
- case anchor
- Lane
- oversight report
Use for oversight context about agency use and inventory management, not proof that a named procurement is active or compliant.
Next: Keep source-linked timeline and role claims limited to the extracted facts.169862e92626