Description: The internet is shifting from a human-first web to an agentic ecosystem, creating new privacy gaps where agents need deep access to personal context. Static, binary security rules ("YOLO risk") are insufficient for this level of autonomy as dynamic systems can change behavior constantly. This workshop introduces Contextual Integrity (CI) as a dynamic normative framework that ensures the appropriate flow of personal data based on social norms. CI acts as both a curriculum for teaching agents to recognize norms and a benchmark for measuring their effectiveness in handling data appropriately across diverse contexts.

Join us for a highly interactive session where we’ll dive right into real-life scenarios! You will have the opportunity to actively brainstorm and work through practical use cases, identifying the pitfalls of existing rules and directly applying CI principles. Most importantly, we will outline concrete ways for you to get involved in the ongoing work around Contextual Integrity, giving you the chance to help shape these emerging standards and enhance the ecosystem's security and privacy for all.

Proposed Discussion Questions:

• Q1: How can we define "correct" behavior in dynamic agentic systems where behavior cannot be pre-defined for every scenario?

• Q2: What are the risks of relying on static, identity-only access controls as agents scale, and how does "YOLO risk" manifest?

• Q3: How can agents establish provable trust through attestation and behavioral constraints rather than blind faith?

• Q4: How do we ensure agentic ecosystems align with human expectations, diverse cultural contexts, and societal norms?