As 2025 draws to a close, one of the most significant changes in federal environmental governance that is least discussed publicly is the quiet expansion of artificial intelligence (AI) behind the scenes across multiple federal agencies. AI tools in federal science programs are not new, but 2025 marked a turning point. Government agencies have begun integrating machine learning models into their daily workflows for exposure modeling, surveillance, targeted enforcement, and environmental monitoring. The White House Executive Order (EO) “Removing Barriers to U.S. Leadership in Artificial Intelligence” dated January 23, 2025, was one of the first EOs for 2025 signed by President Trump. That year was booked by someone else. On Thursday, December 11, 2025, President Trump issued an executive order, “Securing a National Policy Framework on Artificial Intelligence,'' seeking to rein in state-level AI regulation by asserting federal preemption, directing agencies to challenge or block state AI laws, and pressuring Congress to establish a uniform national framework. This move once again demonstrates the current administration's level of commitment to advances in AI.
This decentralized expansion is more pragmatic than ideological, reflecting operational pressures, staffing constraints, and the growing amount of data that agencies must accumulate, evaluate, and summarize. However, the pace of adoption is now outpacing the development of clear policy guardrails to ensure fair and accurate AI-infused products. As we move to 2026the gap between the use of AI and the monitoring of AI is becoming increasingly visible to regulated organizations.
The U.S. Environmental Protection Agency (EPA) continues to expand the use of computing tools within the Office of Research and Development (ORD) and the Office of Chemical Safety and Pollution Prevention (OCSPP) and outlined EPA's own AI compliance plan in response to OMB Memorandum M-25-21. Machine learning-enabled tools such as Open (Quantitative) Structure-activity/property Relationship App (OPERA) and updated read-across algorithms have played a more important role in assessing screening levels and identifying data gaps. EPA also continues to modernize its ToxCast/Tox21 computational toxicology system, which increasingly incorporates statistical and machine learning components to support hazard prediction. EPA's Pesticide Program also describes how AI tools can help meet decision-making timeline goals for pesticide registration applications.
In an enforcement context, EPA regional offices have experimented with data-driven approaches to prioritize testing and identify anomalies in emissions and waste disposal reports. Although these systems are not definitive and EPA emphasizes that inspectors and scientists make the final enforcement decisions, the early use of machine learning-enhanced triage tools suggests that data-driven targeting will continue to expand.
From a policy perspective, these developments highlight two recurring questions for stakeholders.
- If AI or machine learning (ML) output affects assessment or inspection priorities, how will EPA document those effects in the administrative record?
- What opportunities should regulated entities have to understand, replicate, or refute the underlying model?
EPA has not yet issued comprehensive guidance on these issues. This growing gap leads to uncertainty about how these tools intersect with statutory transparency requirements under the Toxic Substances Control Act (TSCA), the Clean Air Act, and other programs, and how their application can be legally supported in the event of a dispute over their legitimacy.
The U.S. Food and Drug Administration's (FDA) Office of Digital Transformation continues to advance AI-enabled tools for pattern detection in large datasets, including tools used to assess contaminants in food-contact materials and identify emerging trends in food safety risks. FDA is piloting AI tools to optimize performance and accelerate drug review protocols, reducing the time needed to summarize adverse events, compare labels, and generate code for database development. The FDA is currently using this AI program, Elsa, to streamline clinical protocol reviews, reduce the time required for scientific evaluation, and identify high-priority test targets. These tools allow FDA to more consistently apply risk-based principles while managing large workloads. Despite these advances, the FDA, like the EPA, has yet to define how insights generated by AI will be incorporated into regulatory decision-making. Stakeholders continue to seek clarity about the role these models play in guiding inspections, enforcement actions, or premarket evaluations.
Beyond the EPA and FDA, other government agencies are also expanding their use of AI in ways that impact health and safety oversight.
- United States Department of Agriculture (USDA) It relies on AI and satellite-based ML models to assess crop health, detect land use change, and support wildfire risk prediction. These functions overlap with climate, conservation, and compliance programs.
- U.S. Department of Energy (DOE) uses AI for grid optimization, materials research, and energy efficiency modeling. AI-enabled predictive tools can inform grant allocation and infrastructure planning decisions. and
- U.S. Department of the Interior (DOI) and U.S. Department of Transportation (DOT) We tested AI tools for habitat mapping, transportation risk modeling, and pipeline integrity assessment.
Across government agencies, most deployments use federally developed or open-source ML frameworks rather than proprietary commercial platforms, reflecting both procurement constraints and the need for transparency. A consistent theme emerged in 2025. That means the adoption of AI is accelerating faster than the policy infrastructure needed to support it. Three gaps stand out, each with legal implications.
- Transparency and reproducibility: Regulators are increasingly relying on models that appear technically sophisticated but are not accompanied by clear documentation of data inputs, assumptions, or uncertainty factors. For regulated entities, this raises questions about how insights generated by AI can be reproduced, authenticated, evaluated, or contested.
- Administrative records integration: If AI tools impact screening decisions, prioritization, or assessment outcomes, agencies will need consistent protocols to document those impacts in administrative records. Without this, legal issues may arise for both administrative and judicial review.
- Consistency between institutions: Different government agencies are deploying AI at different speeds and with different criteria. Without alignment, regulated industries could face a fragmented compliance landscape with varying expectations for data transparency, model validation, and weight of evidence.
Looking to 2026
The story of 2025 is not about AI entering environmental governance group chats. Over the years, its presence and influence in the research field has increased. The big news is that AI has moved into operational decision-making without a corresponding evolution in the development of consistent and transparent governance policies. As government agencies expand adoption of AI, 2026 Additionally, regulated entities will increasingly seek clarity on how these tools impact compliance obligations, enforcement priorities, and risk assessments.
Clear guidance on documentation, transparency, model validation, and reproducibility will help ensure confidence that AI can improve regulatory efficiency without compromising predictability or procedural fairness. Until then, the adoption of AI in health and safety and risk assessment will continue to have a strong but unevenly managed presence in the regulatory environment.
