Johns Hopkins endorses agentic AI for automated toxicological risk assessments

Overview

The field of toxicological risk assessment is undergoing its most profound technological transformation in decades. Researchers at the Johns Hopkins Bloomberg School of Public Health, specifically within the Center for Alternatives to Animal Testing in 2024, have endorsed and demonstrated the application of agentic artificial intelligence to automate complex, multi-step toxicological risk assessments. Historically, evaluating the hazards of chemical substances has been a highly manual, labour-intensive process requiring senior toxicologists to spend weeks or even months combing through literature, synthesising disparate data, and calculating exposure thresholds. By transitioning from traditional, passive large language models to active, autonomous agentic workflows, the environmental and chemical safety sectors can now generate highly accurate, reproducible, and fully auditable risk profiles in a fraction of the time.

For Australian environmental professionals, developers, infrastructure planners, and legal counsels, this development represents a major shift in how contaminated land and chemical liabilities are managed. Site investigations and property transactions frequently encounter delays due to uncertainties surrounding emerging contaminants or complex chemical mixtures that lack established national guidelines. This autonomous technology promises to streamline the preliminary stages of risk characterisation, offering rapid, defensible toxicological profiles that can accelerate due diligence and planning approvals. By understanding this shift, Australian practitioners can better position themselves to harness advanced computational tools while maintaining the rigorous standards required by local regulatory bodies.

The endorsement of agentic systems by a leading institution like Johns Hopkins University signals a departure from simple automated database searching towards true cognitive automation. Unlike basic search scripts or standard generative models, agentic systems possess the ability to plan, use tools, self-correct, and make reasoned judgements regarding study quality and data relevance. This development is particularly timely as global chemical inventories expand rapidly, far outpacing the capacity of traditional manual toxicological evaluation. As regulatory scrutiny intensifies around emerging contaminants, the ability to rapidly assess chemical hazards with high precision is becoming a core capability for modern environmental consultancy and risk management.

Key details

The core of the research endorsed by the Johns Hopkins Bloomberg School of Public Health lies in the distinction between standard generative artificial intelligence and agentic workflows. While standard large language models operate as text predictors, agentic systems are designed with reasoning engines that allow them to plan, select appropriate external tools, execute sequential tasks, and critically evaluate their own outputs. In the context of toxicological risk assessment, the researchers configured a multi-agent system where individual digital agents are assigned specialised roles. For example, one agent is responsible for retrieving literature from primary sources such as PubMed, PubChem, and the United States Environmental Protection Agency CompTox Chemicals Dashboard. Another agent acts as a quality controller, evaluating the methodological strength of the retrieved studies, while a third agent executes mathematical calculations to determine toxicological endpoints.

Crucially, this system addresses the primary challenge of using artificial intelligence in regulatory environments: the black box and hallucination problems. Standard models often generate plausible-sounding but entirely fabricated citations or data points. The agentic workflow developed by the researchers mitigates this by utilising Retrieval-Augmented Generation coupled with mandatory chain-of-thought reasoning logs. Every toxicological claim, dose-response relationship, or hazard classification generated by the system is mapped directly to a traceable source document, complete with Digital Object Identifiers or PubMed IDs. If the system encounters conflicting data, it does not average the results blindly; instead, it executes a pre-defined weight-of-evidence framework to resolve discrepancies based on study quality, sample size, and peer-review status.

The efficiency gains demonstrated by this technology are substantial. Manual preparation of a comprehensive chemical hazard assessment can take an experienced toxicologist between 80 to 120 hours of focused work, spread over several weeks or months. The agentic system can execute the identical data-gathering, screening, synthesis, and reporting tasks in under an hour, maintaining high concordance with assessments conducted by human expert panels. Furthermore, because the system can be updated continuously as new literature is published, the risk profiles remain dynamic rather than static snapshots in time. This capability allows for the rapid screening of tens of thousands of data points, making it feasible to evaluate complex chemical mixtures and novel synthetic compounds that have previously escaped comprehensive toxicological characterisation.

The scientific methodology behind these agents involves a continuous feedback loop. The primary retriever agent extracts data regarding chemical structures, in vitro assays, and in vivo studies. The data is then structured into a standardised format and passed to the analysis agent, which calculates toxicological thresholds such as Benchmark Dose values and No Observed Adverse Effect Levels. A validation agent then reviews the calculations against established international guidelines, flag-marking any inconsistencies for human review. This multi-layered architecture ensures that the final output is not merely a summary of text, but a rigorous, mathematically verified toxicological assessment suitable for regulatory review.

References and related sources

• Primary source: vertexaisearch.cloud.google.com

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.