BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//hacksw/handcal//NONSGML v1.0//EN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20260520T215729Z
DESCRIPTION:Click for Latest Location Information: http://dgiq-edw2026.data
 versity.net/sessionPop.cfm?confid=165&proposalid=16481\nAs organizations de
 ploy multi-agent systems and AI copilots across analytics, operations, and 
 decision support, a new challenge emerges: agents with high-level permissio
 ns interacting with users or agents who have significantly lower privileges
 . This privilege asymmetry can lead to unintentional leakage of sensitive d
 ata, even when traditional access controls are in place.\n\nThis session in
 troduces Asymmetric Access Safety, a governance and architectural framework
  for enabling powerful AI agents while enforcing strict permission boundari
 es. We’ll examine why LLM-driven systems are uniquely vulnerable, how
  “reasoning leaks” occur even without direct access, and what i
 t means to align agent behavior with least-privilege principles. Attendees 
 will learn practical patterns, architectures, and controls for building saf
 e, compliant, and capable multi-agent ecosystems.\n\nWe’ll explore re
 al-world examples, failure modes, mitigation techniques, and design pattern
 s that can enable enterprise-ready AI systems.\n\nKey takeaways: \n\n
 Why multi-agent and human-in-the-loop systems create privilege asymmetry\n
 How LLMs leak data through reasoning, inference, and context propagation\n
 The importance of a strong foundational Data Management\n
 Governance patterns for enforcing Asymmetric Access Safety\n
 Architectural control, including role and session-bound access, guardrails\
 n
 Reference patterns for building compliant analytics agents and copilots\n\n
DTSTART:20260506T114500
SUMMARY:Asymmetric Access Safety: Governing AI Agents Across Permission Bou
 ndaries
DTEND:20260506T122959
LOCATION: See Description
END:VEVENT
END:VCALENDAR