Millie Rittinghouse

Millie Rittinghouse is a Security Data Scientist at Salesforce’s Cybersecurity Operations Center (CSOC). She works on behavioral threat detection for the Agentforce agentic AI platform. Her title at the time of the talk was described as “security database scientist” (transcript), which maps most closely to Security Data Scientist.

At [[unprompted-conference-march-2026|[un]prompted March 2026]], she co-presented “1.8M Prompts, 30 Alerts: Hunting Abuse in a User-Defined Agent Ecosystem” alongside Matt Rittinghouse. Her portions of the talk covered the threat landscape (platform-target attacks vs. abuse of legitimate agency), the limitations of content-moderation-only defenses (reasoning vs. execution blindness; the blocking dilemma; post-generation blindness), and the roadmap toward real-time auto-containment.

Contributions to the wiki

• Execution-layer defense framing — articulated why defenses must operate at the execution layer (what the agent does), not the reasoning layer (what the agent says), in a privacy-preserving multi-tenant context.
• Abuse of legitimate agency — named and defined this threat class as distinct from platform-target attacks: valid, authorized capabilities used maliciously in context.
• Auto-containment roadmap — described the architectural path from batch detection to hot-path inline scoring to automated session kill / token revoke / bot-level lockdown.

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.” If Millie’s last name is Huang, this page should be moved to millie-huang.md. Verify when external confirmation is available.