Agentic AI systems are evolving from passive assistants into autonomous actors that operate across applications, services, and devices on users’ behalf. To function, they often require persistent access to personal data, communications, credentials, and system environments. This challenges privacy and security models built around user-initiated actions and limited data flows, raising questions about encryption, data minimisation, delegated authority and accountability. This workshop will examine how privacy, security, and agency can remain central in the agentic AI era. Rather than adopting an alarmist framing, it will focus on responsible innovation: clarifying how agentic systems alter traditional assumptions and identifying mechanisms to strengthen user control, including scoped permissions, revocable delegation, strong authentication, and oversight. Drawing on lessons from encrypted communications and browser security models, the session will explore how core privacy and security principles can be operationalised in agent architectures, placing user control at the centre of AI agent design and deployment.