Artificial intelligence is becoming increasingly integrated into toxicology and chemical risk assessment. Machine learning, deep learning, large language models, AI-supported QSAR and read-across, PBK modelling, systematic review, evidence integration and decision-support tools are opening new possibilities for more efficient, data-driven and human-relevant safety assessment.
This development is highly relevant to the work of ONTOX, which has explored how advanced computational approaches, artificial intelligence, mechanistic knowledge and New Approach Methodologies (NAMs) can be combined to support next-generation risk assessment. As such technologies move closer to practical application, however, an increasingly important question needs to be addressed:
When can an AI-supported result be trusted sufficiently to contribute to a chemical safety decision?
A new OECD initiative aims to help provide the answer.
At a kick-off meeting held at the OECD in Paris, experts launched the project “High-Level Guidance for the Integration of Artificial Intelligence in Chemical Safety Assessment – Framework for Trustworthy AI in Chemical Safety.”
The project aims to translate the broader OECD AI Principles into practical guidance specifically applicable to chemical safety assessment. The objective is to establish a common baseline for the appropriate use of AI, define expectations regarding human oversight and develop criteria for determining when AI-based NAMs and other AI-supported approaches are sufficiently mature and reliable for a defined regulatory purpose.
Importantly, the ambition extends beyond the development of general principles. The project will explore how concepts such as transparency, explainability, reliability, data quality, uncertainty, traceability and human oversight can be translated into practical criteria, operational procedures, reporting approaches and potentially quality-assurance-style checklists that regulators, developers and users can apply in practice.
Case studies addressing different types of AI use in chemical safety assessment are also expected to form an important parallel workstream, providing an opportunity to test how the proposed principles can be operationalised across different regulatory and scientific contexts.
Strong links with the ONTOX experience
The initiative closely connects with several areas explored during the ONTOX project. ONTOX has worked at the interface between biological and toxicological knowledge, in vitro NAMs, computational modelling, evidence integration and artificial intelligence, while simultaneously considering how emerging approaches can become usable within risk assessment. AI-supported platforms and computational resources applied within the wider ONTOX ecosystem Link to AI tools , demonstrate both the opportunities offered by increasingly sophisticated computational tools and the importance of establishing transparent frameworks for evaluating their reliability, applicability and uncertainty. The OECD initiative therefore represents an important next step: moving from AI innovation toward AI qualification and regulatory trust.
For projects such as ONTOX, this transition is particularly important. Scientific performance alone is not sufficient for regulatory implementation. Developers and regulators also need clarity regarding the intended context of use, data provenance, model performance, uncertainty, reproducibility, transparency, human oversight and the evidence required to demonstrate that an AI-supported approach is fit for its intended purpose.
International collaboration
The OECD project is jointly led by Italy, represented by the Istituto Superiore di Sanità (ISS), and Slovakia, represented by the Centre of Experimental Medicine of the Slovak Academy of Sciences (CEM SAS). Dr Helena Kandárová, Deputy Coordinator of ONTOX and leader of its activities related to risk assessment, implementation and communication, is contributing to the initiative together with Dr. Olga Tcheremenskaia and Dr. Cristina Parenti from ISS, bringing complementary expertise in regulatory toxicology, QSARs, NAMs, validation, risk assessment and the translation of emerging technologies into regulatory practice.
The kick-off meeting brought together an international expert group of more than 60 participants and initiated discussions on the scope of the guidance, organisation of the work and development of the first workstreams. The project is coordinated by the OECD Secretariat, including Dr. Ester Carregal Romero and Dr. Patience Browne. The work will also seek coherence with developments across major regulatory organisations and authorities, including EFSA, ECHA, EMA, FDA and the US EPA, recognising that international alignment will be essential if AI-supported approaches are to become broadly usable in chemical safety assessment.
Representatives (from left): Dr. Helena Kandárová, Dr. Cristina Parenti, Dr. Olga Tcheremenskaia
From innovation to implementation
One of the central lessons emerging from ONTOX is that the successful transition toward next-generation, animal-free safety assessment requires more than developing innovative methods. New approaches must also become understandable, assessable, reproducible and trusted by their intended users.
Artificial intelligence has the potential to accelerate evidence generation and integration considerably. At the same time, its increasing influence on scientific and regulatory decision-making makes robust governance, quality assurance and transparent reporting essential.
The new OECD project provides an important international platform for addressing these challenges and for helping ensure that AI is introduced into chemical safety assessment in a scientifically robust, transparent and responsible manner.
For ONTOX, this initiative also represents an important continuation of the project’s broader objective: moving innovative NAMs and computational approaches from scientific development toward practical implementation in next-generation risk assessment.
You must be logged in to post a comment.