Researchers from Stevens Institute of Technology and Harvard University have developed a new artificial intelligence model that can rapidly predict the potential toxicity of hundreds of previously unregulated chemical byproducts formed during the disinfection of drinking water. Traditional water treatment chemicals such as chlorine and chloramine react with natural organic matter in source water, creating disinfection byproducts (DBPs), some of which have been linked in past research to health issues like bladder cancer and impaired fetal development. The Environmental Protection Agency currently regulates only a small fraction of these byproducts, but the new AI model was trained on existing toxicity data from over 200 known chemicals and used to predict toxicity for more than 1,000 additional compounds, some of which the model suggests could be more toxic than those currently regulated. While experts emphasize that this does not mean tap water is unsafe to drink, the technology gives scientists and regulators a more comprehensive view of potential risks and could inform future monitoring and regulation to protect public health. The work was described in a paper published in the Environmental Science & Technology Letters.
Sources:
https://www.semafor.com/article/01/16/2026/new-ai-seeks-to-make-water-safer
https://www.technologynetworks.com/applied-sciences/news/ai-model-helps-identify-potential-toxic-byproducts-of-disinfecting-drinking-water-408687
https://www.stevens.edu/news/disinfecting-drinking-water-produces-potentially-toxic-byproducts-new-ai
Key Takeaways
- AI is being applied to environmental chemistry to screen and predict toxicity of water treatment byproducts far more rapidly and at far greater scale than traditional lab testing.
- Current regulatory frameworks cover only a fraction of the potentially toxic disinfection byproducts; the AI model highlights compounds that may warrant closer scrutiny and possible future regulation.
- Experts stress that treated drinking water remains safe for consumption, and the goal of the research is to enhance safety through better scientific understanding and potential regulatory improvements.
In-Depth
Artificial intelligence is increasingly being deployed in fields far beyond language and image recognition, and a new application in environmental health could significantly reshape how we safeguard one of the most essential human needs: clean drinking water. Researchers from Stevens Institute of Technology, in collaboration with scientists from Harvard University’s public health school, have developed a machine learning model capable of rapidly assessing the toxicity of the chemical byproducts formed during the disinfection of drinking water. This work was recently highlighted in a technology news report and is based on a research paper published in the Environmental Science & Technology Letters.
Disinfection of drinking water has been one of the most important public health advances in history. By eliminating waterborne pathogens that can cause diseases like cholera, typhoid, and dysentery, chemical disinfectants such as chlorine and chloramine have dramatically reduced illness and death across the globe. However, when these disinfectants interact with organic matter naturally present in water sources, they produce disinfection byproducts (DBPs). Some of these byproducts, like trihalomethanes and haloacetic acids, have been studied in the past and linked to an increased risk of certain cancers and reproductive issues. Despite these known risks, regulatory agencies such as the Environmental Protection Agency only monitor and regulate a small subset of these compounds.
The innovation at the heart of the new research lies in using artificial intelligence to look beyond the limited set of chemicals that have been experimentally tested. Traditional toxicity testing in laboratories is slow, expensive, and labor-intensive, and it is impractical to assess the thousands of potential disinfection byproducts that could form under different conditions. To overcome this constraint, the research team compiled existing toxicity data for more than 200 well-studied chemicals and trained an AI model to learn the relationships between chemical structure and toxicological outcomes. Once trained, the model was able to predict the potential toxicity of over 1,100 other related compounds.
The results were eye-opening. The AI model identified several byproduct compounds with predicted toxicity levels that rival or exceed those of chemicals currently regulated by the EPA. That doesn’t mean that your everyday glass of tap water is suddenly dangerous—experts stress that water utilities are already providing safe drinking water—but it does indicate that our current understanding of DBPs is incomplete. By flagging potentially harmful compounds early in the research and regulatory process, scientists can better target laboratory testing and regulators can consider whether existing safety standards should be expanded or updated.
For instance, the AI-driven predictions could help public health officials focus on specific subgroups of DBPs that warrant detailed toxicology studies. If some of these compounds are confirmed to pose significant health risks, utilities and policymakers might think about adapting treatment processes or setting new regulatory limits. The research also opens the door for future midstream monitoring systems that use similar predictive models to provide real-time risk assessments and early warnings when water quality deviates from safe norms.
From a conservative public policy perspective, there is clear value in embracing pragmatic, evidence-based tools that protect citizens’ health without overregulating or creating unnecessary alarm. Water treatment remains one of the great public health success stories, and innovations like AI-powered toxicity screening can help ensure that benefits remain strong while systematically identifying and mitigating residual risks. Rather than displacing traditional science, the AI model accelerates it—bringing powerful computational tools to bear on problems that have historically lagged due to logistical constraints.
In practical terms, household water remains safe, and basic precautions like proper filtration can further reduce exposure to any trace contaminants. But at the policy and regulatory level, the ability to predict and prioritize potentially harmful compounds more rapidly could help agencies like the EPA make more informed decisions rooted in the best available technology. As AI continues to evolve, its integration into environmental health research promises to sharpen our understanding of complex chemical processes and help maintain safe, reliable drinking water for communities nationwide.