- 7th Annual
- November 17-18, 2026
- Austin, TX
Agent Safety and Security
About the track
This is for AI and ML engineers, architects, and infra leads taking agents into production, plus the directors and VPs who sign off on what those agents are allowed to do.
We’ll cover:
- Agent identity and least-privilege access to tools and data
- Sandboxing and limiting the blast radius when an agent misbehaves
- Red teaming and evals that find failures before launch
- Runtime guardrails and monitoring that catch problems after launch
- Audit trails and human oversight that hold up in regulated industries
Coming from Practitioners who have shipped agents and been burned by them, sharing what broke and what worked, not vendor pitches.
Attendees will walk away with a concrete checklist of controls to put in place before an agent gets production access.
Most agent failures in production aren’t model failures. They’re permission failures: tools and credentials handed to an agent with no real limit on what it can do if something goes wrong. Prompt injection alone now touches six of the ten categories in OWASP’s Top 10 for Agentic Applications, and a 2026 survey of 160 CISOs found 73% call AI agent security a top concern while only 30% say their safeguards are mature. This track is for AI and ML engineers, architects, and infra leads taking agents into production, and for the directors and VPs who sign off on what those agents are allowed to do.
This talk covers agent identity and least-privilege access, plus sandboxing to limit blast radius when something goes wrong. It covers red teaming and evals before launch, and runtime guardrails and monitoring after launch. It also covers the audit trails and human oversight that regulated industries require.
Speakers are practitioners who have shipped agents and been burned by them, not vendors. You’ll walk out with a concrete checklist of controls to put in place before your next agent gets production access.
Track host
Anupama Garani
VP AI Engineering, PIMCO
See Track Lead Profile
Anupama Garani is a Vice President at PIMCO for AI Engineering, with a decade of experience building ML, NLP, and AI systems across Goldman Sachs and PIMCO. She was a core builder of ChatGWM, an enterprise RAG platform on Azure AI featured in a Microsoft customer success story, and placed second in the 2024 PIMCO Innovation Prize.
Independently, she researches why production RAG systems fail in ways standard evaluation misses. She developed the TRUST Framework, a five-dimension diagnostic methodology for enterprise RAG failures, and authored a taxonomy of 33 failure modes across seven pipeline stages, published at ACL’s TrustNLP 2026 workshop. The TRUST Framework has since been adopted independently by graduate researchers at UT Austin and the University of Houston, and by practitioners in construction, financial services, healthcare, and legal tech.
She mentors emerging AI talent through MentisPrep, Break Through Tech, and Topmate, and serves on the TMLS Steering Committee and is a Women in Data Science Ambassador. She has spoken at SSON, AI Camp, AIMUG Austin, UT McCombs, and the American Data Centers & AI Conference, and will speak at ICDMAI 2027 in Bengaluru.
Sessions in this track
MLOps World | GenAI Summit 2026 is a two days of case studies, workshops, and expo on taking AI/ML and agentic systems into production – at the Etter-Harbin Alumni photo – full-bleed hero or browse files.