Matt Rittinghouse

Matt Rittinghouse is a security practitioner at Salesforce’s Cybersecurity Operations Center (CSOC). He works on detecting and responding to threats across the Agentforce agentic AI platform.

At [[unprompted-conference-march-2026|[un]prompted March 2026]], he co-presented “1.8M Prompts, 30 Alerts: Hunting Abuse in a User-Defined Agent Ecosystem” alongside Millie Rittinghouse. His portion of the talk focused on the technical architecture of the behavioral anomaly detection model: the three-level ensemble approach (user / agent / org context), feature design and selection, incremental historical profiling, and the roadmap toward hot-path inline scoring and auto-containment.

Contributions to the wiki

• Three-level ensemble anomaly detection model — first production description of layering user / agent / organization behavioral contexts for agentic AI anomaly detection.
• Feature selection methodology — PCA-inspired approach to culling feature sets; “measure contribution first, then cull to minimum.”
• Deviation-based scoring as confidence proxy — per-axis statistical deviation scores as a confidence-interval analog for alert prioritization.

Name disambiguation

The conference agenda (as captured) lists this speaker pair as “Matt Rittinghouse + Millie Huang.” This page follows the transcript’s file metadata which uses “Millie and Matt Rittinghouse.” Verify when external confirmation is available.