US EPA AI Investigation into Chemical Safety
The decision-making frameworks that govern chemical safety and environmental risk are undergoing a silent, algorithmic shift. Public Employees for Environmental Responsibility (PEER) has launched a formal watchdog investigation into the United States Environmental Protection Agency (US EPA) regarding its reliance on “black box” artificial intelligence and machine learning models to approve new industrial chemicals. This investigation focuses on the regulatory shortcuts taken under the Toxic Substances Control Act (TSCA), where predictive software is increasingly substituted for empirical, laboratory-tested toxicity data.
For Australian environmental professionals, developers, and legal advisors, this development is a critical warning signal. The decisions made by the US EPA do not remain confined within American borders; they form the bedrock of global chemical registries and toxicity databases that Australian regulators rely upon. If the foundational toxicological profiles of these chemicals are being generated by unverified, proprietary algorithms, then the environmental liabilities, due diligence assessments, and remediation standards we apply today may be built on highly unstable scientific foundations.
This transition to automated regulatory approvals introduces a latent layer of risk to commercial property transactions, brownfield developments, and infrastructure projects. When predictive models fail to anticipate the real-world environmental persistence or bioaccumulative nature of a novel compound, the downstream liabilities do not fall on the software developers or the chemical manufacturers. Instead, they land squarely on the landowners, developers, and environmental consultants who must manage the resulting soil, groundwater, or soil vapour contamination years after the asset has been developed.
The Role of QSAR Models in TSCA Approvals
The core of the PEER investigation lies in a detailed Freedom of Information Act (FOIA) request filed with the US EPA Office of Pollution Prevention and Toxics. The watchdog group is seeking comprehensive records detailing how the agency validates, audits, and applies artificial intelligence, machine learning, and predictive models during the assessment of new chemicals under Section 5 of TSCA. Specifically, the inquiry targets the quantitative structure-activity relationship (QSAR) models and other computational tools used to estimate toxicity when physical testing data is completely absent.
QSAR models operate on the scientific assumption that chemicals with similar molecular structures will exhibit comparable physical, chemical, and biological behaviours. While this is a standard screening method in early-stage chemical development, the US EPA has increasingly relied on these algorithmic estimations to grant commercial approvals without requiring manufacturers to submit empirical in vivo or in vitro toxicological data. The PEER investigation seeks to uncover whether these models have been subject to rigorous, independent peer review, what their documented error rates are, and whether the EPA is utilising proprietary algorithms whose source code is shielded from public and scientific scrutiny as trade secrets.
The systemic reliance on predictive toxicology has been accelerated by the administrative pressures introduced by the 2016 amendments to TSCA, which mandated the US EPA to make affirmative safety determinations on all new chemical submissions within strict, statutory timeframes. Confronted with a massive backlog and a lack of resources, the agency expanded its use of computational models to automate and accelerate the review process. The watchdog investigation questions whether this shift represents an unlawful delegation of regulatory authority to unvalidated software, effectively creating an opaque regulatory environment where chemicals are declared safe based on mathematical speculation rather than empirical evidence.
The implications of this investigation are amplified by the scale of chemical introductions. Thousands of new synthetic compounds, including novel solvents, plastics, flame retardants, and surfactant formulations, are reviewed globally each year. If the US EPA predictive models suffer from structural blind spots, they may consistently generate false negatives, classifying highly persistent, bioaccumulative, or toxic substances as low-risk materials. When these substances eventually enter the global market and industrial supply chains, they inevitably contaminate soil and groundwater systems, leading to retroactive regulatory actions and substantial remediation liabilities.
How US EPA Approvals Impact Australian Contaminated Land Guidelines
The regulatory frameworks governing contaminated land and chemical safety in Australia are deeply intertwined with international scientific assessments. Under the Industrial Chemicals Act 2019, the Australian Industrial Chemicals Introduction Scheme (AICIS) regulates the importation and manufacture of industrial chemicals. The AICIS categorisation process relies heavily on assessments conducted by trusted international authorities, with the US EPA TSCA determinations serving as a primary reference point. If a chemical has been cleared for commercial use in the United States, Australian importers can often categorise the introduction as a lower-risk pathway, avoiding the need for extensive domestic laboratory testing. Consequently, any systemic flaws or algorithmic errors within the US EPA reviews are directly and immediately imported into the Australian market.
Furthermore, the National Environment Protection (Assessment of Site Contamination) Measure 1999, commonly referred to as the NEPM 2013, establishes the national guidelines for assessing site contamination in Australia. The Health Investigation Levels (HILs) and Ecological Investigation Levels (EILs) detailed within the NEPM are derived from toxicity reference values established by international bodies, including the US EPA, the World Health Organization, and the European Chemicals Agency. Where these underlying reference values are based on algorithmic predictions rather than verified laboratory data, the trigger levels used to assess contaminated sites across Australia may understate the true risk to human health and the environment.
References and related sources
- Primary source: peer.org
How iEnvi can help
iEnvi provides specialist consulting services relevant to this topic. Our team includes CEnvP Site Contamination Specialists with experience across contaminated land, groundwater, remediation, ecology, and regulatory compliance.
- iEnvi contaminated land investigation services
- iEnvi remediation and validation services
- iEnvi expert services and independent review services
This is an iEnvi Machete news summary. Prepared by iEnvi to summarise the source article for contaminated land, groundwater, remediation, approvals and site risk professionals.
Published: 17 Jun 2026
Need advice on this topic? Speak to an iEnvi expert at info@ienvi.com.au or 1300 043 684, or contact us online.
Need advice on this issue? iEnvi provides practical, senior-led environmental consulting across contaminated land, remediation, ecology and environmental risk.
Team credentials Contaminated land services Remediation services Groundwater services Talk to iEnvi