NEJM AI. 2026 Jan;3(1):AIe2501301. doi: 10.1056/AIe2501301. Epub 2025 Dec 24.ABSTRACTAbstractThe transition from animal testing to new approach methodologies (NAMs) is a scientific and ethical imperative championed by regulators and industry. In this issue, Tieghi and colleagues outline a federal road map to accelerate this shift; however, we argue that the operationalization of NAMs is currently hindered by a lack of benchmarking infrastructure. Drawing parallels to the "AlphaFold moment" in protein structure prediction, we posit that success requires community-driven data standards and neutral evaluations that define structural biology. Currently, predictive toxicology is constrained by undefined phenotypic end points, fragmented data, and reliance on animal models with limited human concordance. To replicate the prediction success of AlphaFold, the field must build a data infrastructure for toxicology - establishing open, standardized repositories and recurring, blinded challenges. Key differences between AlphaFold and NAMs, however, are the risk functions and tasks - one is built for predictive discovery, the other for harm mitigation, which requires integration with regulatory decision-making frameworks. Nevertheless, establishing data infrastructure is the prerequisite for transforming NAMs from niche research tools into trusted regulatory instruments capable of reliably predicting human harm.PMID:42932530 | DOI:10.1056/AIe2501301