As federal agencies accelerate artificial intelligence (AI) adoption, buying the technology is only part of the challenge. Agencies must also ensure that the models they purchase perform as promised, protect sensitive government information, and meet mission requirements. That means moving beyond vendor assurances and putting clear, enforceable expectations into AI contracts.
Recent federal initiatives aim to create an approach that provides guardrails for secure, unbiased AI without slowing innovation.
Executive Order Provides Federal Guidance
An executive order issued in June, “Promoting Advanced Artificial Intelligence Innovation and Security” (EO 14409), outlines the oversight national security and civilian agencies should exercise in purchasing AI-enabled technologies. The order encourages developers of advanced AI models to give the federal government 30 days of access before broader release, as early access to these models gives the government an opportunity to evaluate emerging capabilities and cybersecurity risks.
EO 14409 also directs the government to establish a classified benchmarking process to identify advanced models that meet the definition of “covered frontier models.” Creating a consistent evaluation criteria means the government is not relying solely on vendors’ claims about model capabilities. Agencies also need their own acquisition guardrails, including mission-specific performance testing, data protection requirements, and ongoing monitoring, to determine whether a model is suitable for their particular needs.
Legislative Efforts to Secure AI
The bipartisan draft bill The Great American AI Act aims to create a federal framework for AI governance. The bill would formally codify and fund the Center for AI Standards and Innovation, which would develop voluntary guidelines, best practices, and standards for AI security, as well as evaluate AI systems and monitor AI progress. The bill also calls for the creation of a testbed program where public and private sector labs and entities would collaborate on evaluating AI systems.
While still in draft, this bill speaks to the belief that AI should be tested and measured before government use. Legislators are warning against relying on vendor demonstrations or descriptions to assess risk and model efficacy.
Operational Support From the GSA
The General Services Administration (GSA) has been working on rules for an acquisition regulation clause regarding solicitations and contracts that involve AI capabilities. Over the past several months, the agency has sought input from industry and government to create guardrails that are not overly restrictive.
With the feedback it has received, the GSA has evolved its plans from a broad AI clause to one focused on LLMs processing government data. This revision process provides an example of how, rather than developing guardrails purely through adding a list of rules, agencies need to be clear in their expectations, definitions, and requirements—as well as ensuring those expectations and requirements are enforceable.
To follow the work of AI acquisition oversight and model risk evaluation, check out these events and resources:
- AI Buyer’s Guide for Government (white paper) – This guide explores how federal, defense, intelligence, and state and local agencies are adopting AI technologies, including generative AI, agentic AI, secure AI, AI infrastructure, and advanced analytics. It provides practical insight into emerging federal AI policies, procurement strategies, and real-world implementation success stories from agencies such as the NIH, DoD, VA, and DOE.
- A Roadmap for Success: Building a Comprehensive AI Adoption Playbook (white paper) – This outline provides a robust framework for building an AI adoption playbook that can serve as a practical guide to navigating the complexities of AI integration—from executive buy-in to equitable implementation and scalable growth.
- Artificial Intelligence Acquisitions: Agencies Should Collect and Apply Lessons Learned to Improve Future Procurements (white paper) – This GAO report reviews how agencies acquire AI technologies and identifies common procurement challenges, including defining requirements, protecting data rights, evaluating AI performance, managing costs, and adapting traditional acquisition processes to rapidly evolving technologies.
- Carahsoft’s Enterprise Open Source & AI Forum (November 18, 2026; Reston, VA) – This event will convene leaders and practitioners across government and industry for recent developments in open source, emphasizing commonalities between open-source software and open-source AI models.
- AITalks (April 29, 2027; Washington, DC) – As large-scale generative and agentic AI systems have moved from experimentation to operational deployment, artificial intelligence has become a defining priority for federal agency leaders and U.S. policymakers alike. This event brings together technologists, program executives, and policymakers to discuss how AI is being integrated into government operations.
Search GovWhitePapers and GovEvents to find even more information on the future of AI acquisition reform.
