Artificial Intelligence and Machine Learning in Fragrance Safety Assessment: Current Applications and Future...

Artificial intelligence (AI) and machine learning (ML) are steadily becoming part of routine practice in fragrance safety assessment rather than remaining research curiosities. This manuscript describes how the Research Institute for Fragrance Materials (RIFM) currently applies these methods and identifies where they are most likely to expand. It includes current applications across all the endpoints, including read-across and structural profiling for skin sensitization, genotoxicity, and repeated-dose and reproductive toxicity; metabolism prediction with simulators such as TIMES; inhalation dosimetry; early-stage phototoxicity screening; and environmental fate and ecotoxicity estimation using tools like EPI Suite. A recurring theme is RIFM's shift toward reproducible, model-driven approaches-Structure-Activity Groups and clustering of over 6,000 materials-built on a four-decade safety database. The manuscript then considers emerging tools: generative models, deep learning for hard-to-test endpoints, high-throughput endocrine screening, and language models for exposure analysis and communication. It also weighs the factors that will determine regulatory acceptance: limited model transparency, the need for validation consistent with (Q)SAR principles, uncertainty reporting, and adherence to FAIR data practices. Throughout, the central argument is that AI is most useful alongside expert judgment-broadening the scale and consistency of assessments while preserving human oversight while supporting the wider move to reduce animal testing.