The right to erasure lets individuals request deletion of personal data when applicable legal conditions are met. In agent systems, deletion can be difficult: memories, logs, and learned preferences may persist across components or be passed to other agents and services. section 5.6.3
Treat erasure as a system-wide data-lifecycle control, not just a database operation. Persistent and distributed memory can leave data in downstream services, where it may be retained independently or used for other purposes. Set retention timelines for agent data and apply them to services receiving that data, unless further use is needed. section 5.6.3 section 5.0
Operational risks and controls include:
- Identify where personal data persists, including agent memory, interaction histories, action logs, and downstream agent or service storage. section 5.6.3
- Define retention periods and use automated deletion scripts to clear old agent data and caches. section 5.6.3
- Account for data copied, transformed, or learned by systems, including retrieval indexes, model parameters, vendor systems, and backups. section 5.0
A booking agent
Booking a doctor’s appointment can send health data to calendar, transit, and insurance services. Each may retain it independently, so an erasure process must account for the services the agent touched. A separate retention rule might delete chatbot transcripts older than 60 days. section 5.6.3



