Private-sector advisors who work with government clients – or whose clients sell products and services to governments – have a direct interest in public-sector AI governance. Government AI procurement requirements frequently cascade to vendors: if an agency must demonstrate that its AI tools meet certain standards, the vendor supplying those tools must meet those standards too. Understanding public-sector governance requirements is competitive intelligence for anyone operating in or adjacent to the government market.
Canada: Federal Framework
The Treasury Board of Canada Secretariat’s Directive on Automated Decision-Making, in force since April 2019 and applicable to systems developed or procured after April 2020, governs all federal departments and agencies that use automated decision systems to make or support administrative decisions affecting individuals. It requires:
- Completion, approval, and public posting of an Algorithmic Impact Assessment (AIA) prior to deploying any automated decision system, with the results published on the Open Government Portal
- A four-level impact framework (Level I through Level IV) that escalates transparency, explanation, human oversight, and approval requirements based on the severity of decisions being automated – Level III and IV decisions require the final decision to be made by a human
- Peer review by qualified external experts for medium- and high-impact systems
- Role-based staff training and meaningful recourse mechanisms for individuals affected by automated decisions
The Government of Canada’s Guide on the Use of Generative Artificial Intelligence, published by Treasury Board and updated most recently in June 2025, provides practical guidance to federal public servants on responsible use of generative AI tools. It covers challenges specific to generative AI (hallucination, privacy, intellectual property), principles for responsible use, and best practices. It applies to all federal institutions.
The Schwartz Reisman Institute’s analysis of the post-AIDA landscape concludes that Canada is unlikely to pass comprehensive federal AI legislation for the next several years. The expected governance ecosystem instead relies on: Treasury Board policy instruments (the Directive on Automated Decision-Making, the AIA Tool); provincial regulation (including Ontario’s Bill 194 for the public sector); and sectoral regulators developing AI-specific guidance within their existing authorities.
United States: Federal Framework
The Biden-era OMB Memorandum M-24-10, Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence, issued March 28, 2024, has been superseded. On April 3, 2025, the Trump administration replaced it with M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust, along with a companion procurement memo (M-25-22). The new memo shifts emphasis from precautionary governance to accelerated adoption, but retains several substantive requirements: agencies must designate Chief AI Officers within 60 days; large agencies must convene AI governance boards within 90 days; agencies must at least annually inventory and publicly publish their AI use cases; and for high-impact AI uses (those whose outputs serve as a principal basis for decisions with legal, binding, or significant effects on rights or safety), agencies must conduct pre-deployment testing, complete AI Impact Assessments, ensure human oversight, and offer remedies or appeals for affected individuals.
As of January 2026, the U.S. federal AI use case inventory documented over 2,133 publicly reportable use cases across 41 federal agencies – a figure that is rising as adoption accelerates under the current administration’s deregulatory posture. FedScoop’s reporting on the 2025 inventory cycle confirms that OMB continues to compile and publish the consolidated list on GitHub.
US State and Municipal
California’s Governor Newsom issued Executive Order N-5-26 on March 30, 2026, directing the California Department of Technology to implement new trust and safety requirements on AI companies contracting with California agencies – including certifications about policies protecting public safety, preventing harmful bias, and protecting civil liberties. This builds on Newsom’s 2023 Executive Order N-12-23 on generative AI. Both orders illustrate how state procurement requirements create de facto governance obligations for any vendor seeking state contracts.
Texas’s Responsible Artificial Intelligence Governance Act (TRAIGA), signed in June 2025 and effective January 1, 2026, focuses on prohibiting intentional use of AI to unlawfully discriminate, produce child sexual abuse imagery, or enable government social scoring – a narrower scope than the Colorado AI Act, with enforcement solely through the state Attorney General.
NYC’s algorithmic tools framework, anchored by Local Law 144, continues to expand the city’s role as a laboratory for urban AI governance.
Procurement as Governance Vector
When governments require that AI tools meet specific standards before purchase – transparency documentation, bias audits, human override capabilities, incident reporting – those requirements flow directly to vendors. Organizations seeking government contracts should treat public-sector AI governance requirements as a compliance prerequisite, not a back-of-mind consideration. The GSA’s AI procurement guidance and Canada’s evolving vendor qualification process (referenced in the Schwartz Reisman Institute analysis) signal that AI governance documentation will increasingly be a condition of doing business with government, not merely a reputational asset.
