The attack is already in your index
Panel · Day 2 · Tue, Nov 10
Why retrieved documents, tool responses and memory are the softest way in
Session
About this session
In 2025, researchers embedded instructions in an ordinary web form and watched a Salesforce Agentforce agent retrieve sensitive CRM records and send them to an external address. Nothing about the model was compromised; the payload arrived the way every other piece of context arrives. This panel covers what teams are actually putting in place: provenance tracking that traces an action back to the document that triggered it, validation of external inputs before an agent acts on them, and detection that works on content nobody wrote by hand. Expect disagreement on whether poisoned context can be reliably cleaned, or whether untrusted content must simply stay away from anything privileged.
Agenda
Also on Day 2 · Tue, Nov 10
All times PT.
- 9:15 – 10:00 AMHow do agents work, and how should they work?Panel
- 10:15 – 11:00 AMEvery sandbox ends up with exceptionsPanel
- 11:15 AM – 12:00 PMBuild secure always-on local agents on AMD Ryzen AI PC using LemonadeTutorial
- 1:15 – 2:00 PMMaking a data agent trustworthy with a semantic layer and auditable memoryTutorial
- 2:15 – 3:00 PMTeaching coding agents to think in GPUTutorial
Registration
Get notified for the next conference.
Registrations are live for the March 30–31, 2027 edition of the Agentic AI Conference.
Register with email
You're on the list. Watch your inbox for your Zoom link. See you November 9.


