AI memory is stored context or historical state that an AI system uses to shape later responses. It may include conversation history, user preferences, summaries, or application-specific records, and can be stored in a database or managed by an AI platform. Memory is distinct from the model itself: changing or replacing a model does not necessarily move the information or reproduce the same behavior.
Memory matters for security because it can contain sensitive information and affect what the system reveals or does. In attacks, a system that retrieves untrusted or maliciously crafted stored content may be manipulated into following unsafe instructions or disclosing data. Defenders should control what is saved, apply access and retention rules, and test retrieval behavior. When migrating providers, inventory and export memory schemas and dependencies, then validate that privacy controls and critical workflows still work in the new environment.