Persistent memory lets AI agents treat past interactions as durable state, but current systems give little control over whether that state remains valid or should be used in a later query. A new arXiv paper argues that reliable memory requires explicit governance of both the lifecycle of stored state and its eligibility at query time.
The authors propose PLACE, a framework designed to manage these two aspects: how memory state evolves over time and whether a given memory should influence a particular query. The abstract notes that existing systems provide limited control over these decisions, and that reliable memory depends on addressing them directly.
While the full details of PLACE are not in the abstract, the core claim is clear: without lifecycle and query-time eligibility rules, persistent memory can become stale or inappropriately influential. The paper positions PLACE as a step toward more dependable memory in agent systems.