The use of artificial intelligence in agency decision-making across the federal government is a cornerstone of the Trump administration’s effort to drive efficiency. The combination of large amounts of information, the need to evaluate that information for quality and relevance, and the potential for significant public interest have contributed to the U.S. Environmental Protection Agency (EPA) announcing plans to deploy AI. To date, actual implementation of AI appears to be lagging behind EPA’s stated intentions, but there are instances in which the Agency is applying or plans to apply AI in connection with its regulatory functions. Stakeholders should be aware of what EPA has done to date and its future plans.  

Trump Administration AI Policy: Federal Agency Requirements and the National AI Action Plan

The Trump administration has made adoption of AI within the federal government a key component of its overall efforts to enhance government efficiency. As early as two weeks after the president’s inauguration, the Office of Science and Technology Policy (OSTP) and Office of Management and Budget (OMB) required agencies to develop AI strategies and sought public input on priority actions to include in the AI Action Plan. This resulted in guidance to agencies in OMB Memoranda M-25-21 and M-25-22, finalized in April 2025. The administration’s approach to AI has continued to evolve. On March 20, 2026, the White House issued its National Policy Framework on Artificial Intelligence. This document sets forth the administration’s blueprint for AI that it hopes will be embodied in legislation. The Framework calls for preemptive federal AI regulation, displacing state activity and focusing on innovation rather than regulation. Additionally, the Framework calls for Congress to ensure that national security agencies have the technical capacity to evaluate and mitigate risks posed by evolving AI models.

EPA’s AI Strategy and Compliance Plan: Use Cases and Regulatory Applications

Adoption of AI by regulatory agencies presents challenges separate from the AI governance set forth in overall policy guidance. In response to the April 2025 OMB memoranda, EPA issued its AI Compliance Plan and AI Strategy last October. These documents impose controls on the use of AI and identify multiple potential use cases. In public statements, EPA has indicated that it views its workload as uniquely well suited to AI assistance. In its October 2025 Final AI Strategy, EPA identifies 18 current AI use cases and proposes uses it is currently investigating, including AI screening of scientific studies for quality based on pre-set criteria and speeding pesticide applications by having AI generate summaries and evaluations of certain required studies.

EPA’s 2025 AI Use Case Inventory: How Far Has Implementation Actually Advanced?

OMB guidance requires EPA and other agencies to publish annual AI Use Case compilations. EPA’s AI Use Case Inventory for 2025 was published in February 2026 and updated in April 2026.  While some press reports emphasize the expanding number of uses, digging into the information reveals that practical adoption of AI by EPA is much more aspirational than real.

The 2025 inventory includes a total of 82 specific items (based on EPA’s sequential numbering), although not all are publicly available. The spreadsheet appears to reflect a cumulative compilation of all deployed, pilot, pre‑deployment, and retired use cases. According to EPA’s Consolidated Summary document, many current use cases of commercial AI tools are low-level and commonplace, such as for scheduling, comparing documents, or deploying helpdesk chat assistants for technical support. EPA classifies only one deployed and one pre-deployment use case as high-impact, and one use case as “presumed high-impact but not determined high-impact.” 

EPA’s High-Impact AI Use Cases: RCRA Enforcement, Lead Abatement, and Surveillance

The only deployed high-impact use case (item 78 on EPA’s table) involves using AI to prioritize RCRA inspection and enforcement for Large Quantity Hazardous Waste Generators. The identified benefits include reduced staff time and improved identification of potential violators. The model was trained on historical compliance data. 

EPA’s pre-deployment high-impact use case (item 43 on the table) focuses on EPA’s lead abatement program by reviewing documents, photos, or videos for TSCA renovation, repair, and painting work descriptions as well as the review of leases for lead disclosure rule violations. EPA has stated its intent is to further use AI to draft enforcement actions “with the understanding that the final determination of compliance and environment actions is inherently a government function.”

The “presumed high impact” use case is item number 46, “Brief Cam.” This AI tool is being deployed to review surveillance camera records to highlight segments and bookmark still images for review by EPA Special Agents. It is currently characterized as a pilot and intended to be used in law enforcement investigations.

Other Notable EPA AI Applications: Public Comments, Pesticide Registration, NAAQS, and Clean Air Act Review

While not classified as high impact, several other identified use cases also are noteworthy, including:

• Use 57: Pilot use of Generative AI by EPA Region 8 to summarize public comments and prepare draft responses. EPA explains this is not the “principal basis” for decisions and therefore is not high impact.
• Use 60: Pre-deployment use of AI to extract data from pesticide registration-related documents to allow comparability across submissions and consistency.
• Use 66: Pre-deployment use of Natural Language Processing (NLP) AI to process public comments on proposed rules and identify “substantive material.” 
• Use 76: Deployed Machine Learning (ML) AI to rank and prioritize scientific literature for review in connection with Clean Air Act NAAQS revision. 

Although not considered high impact by EPA, these AI applications can carry direct regulatory consequences and it is important to monitor the development and evolution of these activities going forward. 

AI and the Administrative Procedure Act: Legal and Regulatory Compliance Considerations

AI use in regulatory settings such as these has drawn attention from legal scholars. In February 2026, the Yale Journal of Regulation published a symposium on AI and the Administrative Procedure Act (APA). The symposium articles address both potential advantages and concerns associated with agencies use of AI in making regulatory decisions. One article argues that the arrival of AI presents an opportunity to reshape the administrative state, raising questions about whether agencies must disclose details about their algorithms when relying on them in promulgating rules, drawing on the Loper Bright doctrine’s requirement that agencies disclose certain facts on which they rely in promulgating a rule. Another article examines what it means for an agency to “use AI,” given the range of ways that large language models can be utilized at every stage of policy formulation, and notes that how administrative law adapts to AI will depend on the specific facts of each situation — particularly whether AI use touches on rights or safety-impacting domains.

Of relevance is a discussion of whether having “a human in the loop” represents adequate oversight of AI and whether such an approach would satisfy the APA’s requirements. One perspective, expressed by The Ohio State University Assistant Professor of Law Bridget C.E. Dooling, is:

We kid ourselves if we think that a “human in the loop” is more than an impoverished way to think about what agencies owe the public. We can likely make great progress in regulatory policy by letting algorithms into our loop, not the other way around.

A question for EPA and stakeholders is evaluating the extent to which AI can make, or at least inform, decisions that can meet the APA standard and withstand judicial scrutiny.

Conclusion

Despite attention paid to AI adoption at EPA, meaningful deployment in consequential regulatory decision-making remains limited at this time. The trajectory, however, points toward expanded use, and stakeholders should not be lulled into complacency. As EPA moves beyond administrative functions and into AI-assisted rulemaking, enforcement targeting, and scientific review, legal challenges under the APA may arise — particularly around whether AI-assisted decisions satisfy the requirement for reasoned explanation and whether a “human in the loop” constitutes meaningful oversight. Stakeholders with interests before EPA should be paying attention now, before the technology gets ahead of the legal framework designed to hold it accountable.