Several challenges remain with the regulatory use of the Threshold of Toxicological Concern (TTC), particularly in the context of advances in Artificial Intelligence (AI), New Approach Methodologies (NAMs), and next-generation risk assessment (NGRA).

This presentation examines how these advances are driving an evolution of TTC. Emphasis is placed on AI-assisted leveraging of large toxicological databases to expand endpoint coverage beyond systemic, genotoxic, and neurotoxic effects. AI-driven chemical grouping and local neighborhood approaches are presented as a modern alternative to the traditional Cramer classification. The integration of NAM-derived evidence and physiologically based kinetic (PBK) models for deriving internal TTC (iTTC) values is also explored. Finally, the presentation discusses the transition from deterministic TTC values towards probabilistic expressions of toxicological concern, enabling quantitative characterization of uncertainty and variability while maintaining a high level of human health protection.

These concepts are being implemented as a decision criterion within the ONTOX Probabilistic Risk Assessment (OPRA) framework, developed under the Horizon 2020 ONTOX project. Designed as an AI-enabled system, OPRA integrates probabilistic risk assessment methodologies with computational tools and data sources for rapid, user-friendly, scientifically robust toxicological evaluations. It is computationally integrated within Toxindex, an agentic platform powered by large language models (LLMs), specialized AI agents, and a comprehensive collection of toxicological databases, literature, regulatory reports, and computational tools. ToxIndex retrieves, predicts, processes, and integrates evidence at each stage of the OPRA workflow, applying predefined decision criteria, including those based on TTC. All evidence, intermediate steps, and conclusions are documented in transparent, reproducible reports.