GELATO: Triggering with Anomaly Detection in ATLAS Run 3

Anomaly detection (AD), a set of machine learning methods that tout minimal model dependence, have continued to gain popularity in the ATLAS experiment as we search for new physics. While current AD analyses have produced exciting results, it remains a possibility that such new physics evades the standard set of triggers. Integrating AD into the trigger system therefore provides a unique opportunity to enhance detection efficiency for unknown signatures while maintaining a model agnostic approach. This work presents the ATLAS Generic Event Level Anomalous Trigger Option (GELATO) chain, which is now running across the Level-1 and High-Level triggers during 2025 datataking. We demonstrate performance across a suite of simulated benchmark signals, and present a glimpse of GELATO’s behavior on real collision data.