GateMem: Benchmarking Memory Governance in Multi-Principal Shared-Memory Agents

Memory benchmarks for LLM agents largely assume single-user settings, leaving shared assistants for hospitals, workplaces, campuses, and households understudied. In these deployments, multiple principals write to a common memory pool and query it under different roles, scopes, and relationships, so memory quality requires governance as well as recall. We introduce GATEMEM, a benchmark for multi-principal shared-memory agents. GATEMEM jointly evaluates utility for legitimate long-horizon requests with state updates, access control across contextual authorization boundaries, and agent-facing active forgetting after explicit deletion requests. It spans medical, office, education, and household domains, with long-form multi-party episodes, incremental memory injection, hidden checkpoints, structured judging, and leak-target annotations. Across diverse baselines and backbone models, no method simultaneously achieves strong utility, robust access control, and reliable forgetting. Long-context prompting often yields the best governance score at high token cost, while retrieval-based and external-memory methods reduce cost yet still leak unauthorized or deleted information. These results show current memory agents remain far from reliable shared institutional deployment.

Introduction. Large language model (LLM) agents are increasingly designed as persistent assistants rather than stateless chat systems, with memory mechanisms that maintain, retrieve, and update information across interactions [22, 40, 21, 7]. This shift has made memory central to long-term interaction, adaptation, and personalization, and has made the evaluation of stored context a critical research priority. Recent benchmarks have substantially advanced memory evaluation, covering incremental updates, lifelong learning, and long-horizon reasoning [12, 38, 28, 2]. Yet as LLM agents transition from personal chatbots to persistent institutional assistants, a critical deployment regime remains insufficiently studied, namely the multi-principal shared environment [24, 33]. Figure 1 illustrates this shift from conventional memory evaluation to shared-memory governance. Most existing memory benchmarks treat an agent’s memory as a private cache, where failures primarily affect a single user and the objective is maximum recall.

Discussion / Conclusion. We introduced GATEMEM, a benchmark for evaluating memory governance in multi-principal shared-memory agents across utility, access control, and active forgetting. Our experiments reveal that current agent designs struggle to satisfy these requirements simultaneously. Long-context prompting offers the strongest governance trade-off but at high computational cost, while retrieval and external-memory baselines remain vulnerable to unauthorized disclosure and post-deletion recovery. These results suggest that future agents must treat memory not merely as a recall resource, but as a governed shared state with reliable access and deletion semantics.