Quick Buzz Feed
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.
The Trump administration has made adopting AI within the federal government a core initiative to boost efficiency. The Office of Science and Technology Policy (OSTP) and Office of Management and Budget (OMB) mandated AI strategies from agencies and solicited public input for the AI Action Plan. This led to guidance in OMB Memoranda M-25-21 and M-25-22 in April 2025. Subsequently, the White House issued its National Policy Framework on Artificial Intelligence in March 2026, advocating for preemptive federal AI regulation, prioritizing innovation, displacing state activity, and ensuring national security agencies can assess AI risks.
The adoption of AI by regulatory bodies presents unique challenges beyond general policy guidance. In response to the April 2025 OMB memoranda, EPA released its AI Compliance Plan and AI Strategy in October. The EPA asserts that its extensive workload is particularly suited for AI assistance, outlining 18 existing AI use cases and exploring future applications. These include using AI to screen scientific studies for quality against predefined criteria and to expedite pesticide applications by generating study summaries and evaluations.
OMB directives require EPA and other agencies to release annual AI Use Case compilations. EPA's 2025 inventory, updated in April 2026, lists 82 specific items. Despite reports suggesting an expansion of AI uses, a closer look reveals that practical implementation by EPA is largely aspirational. Many current applications involve common, low-level commercial AI tools for tasks like scheduling, document comparison, or technical support chatbots. The inventory classifies only one deployed and one pre-deployment use case as high-impact, with an additional 'presumed high-impact' case.
The sole deployed high-impact AI use case (item 78) focuses on leveraging AI to prioritize RCRA (Resource Conservation and Recovery Act) inspections and enforcement actions against Large Quantity Hazardous Waste Generators. This aims to reduce staff time and improve the identification of potential violators by training the model on historical compliance data. The pre-deployment high-impact use case (item 43) assists EPA’s lead abatement program by reviewing documents, photos, or videos related to TSCA (Toxic Substances Control Act) renovation work and lead disclosure rule violations in leases. EPA intends to use AI for drafting enforcement actions, with human oversight for final determinations. The 'presumed high impact' use case (item 46), 'Brief Cam,' is a pilot for reviewing surveillance camera footage to flag segments and bookmark images for EPA Special Agents in law enforcement investigations.
Beyond high-impact cases, several other significant AI applications are underway at EPA: Use 57 involves a pilot of Generative AI by EPA Region 8 to summarize public comments and draft responses, though not as the principal basis for decisions. Use 60 is a pre-deployment AI tool designed to extract and standardize data from pesticide registration documents for improved comparability and consistency. Use 66 employs Natural Language Processing (NLP) AI to analyze public comments on proposed rules, identifying substantive material. Finally, Use 76 utilizes deployed Machine Learning (ML) AI to rank and prioritize scientific literature for review in the context of Clean Air Act National Ambient Air Quality Standards (NAAQS) revisions. These applications, while not high-impact according to EPA, carry direct regulatory consequences and warrant close monitoring.
The application of AI in regulatory contexts has sparked considerable debate among legal scholars. A Yale Journal of Regulation symposium explored AI and the Administrative Procedure Act (APA), addressing both the benefits and concerns of AI in regulatory decision-making. Discussions included whether agencies must disclose algorithmic details when promulgating rules, drawing parallels to the Loper Bright doctrine, and how to define 'using AI' given its varied roles in policy formulation. A critical discussion point is the adequacy of 'human in the loop' oversight and its compliance with APA requirements, with some scholars advocating for a more profound integration of algorithms into regulatory processes.
Despite significant discussion surrounding AI adoption at the EPA, its meaningful deployment in crucial regulatory decision-making remains limited for now. However, the trend indicates an expanding role for AI. Stakeholders should not become complacent, as EPA's move towards AI-assisted rulemaking, enforcement targeting, and scientific review is likely to generate legal challenges under the APA. Key issues will revolve around whether AI-supported decisions provide sufficient reasoned explanations and if a 'human in the loop' constitutes adequate oversight.