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

GateMem is a newly introduced benchmark designed to evaluate memory governance in multi-principal shared-memory agents, addressing the shortcomings of existing benchmarks that focus on single-user scenarios. It assesses utility in long-horizon requests, access control, and active forgetting across various domains, including healthcare and education, utilizing long-form multi-party episodes and structured judging. The findings indicate that while long-context prompting can yield high governance scores, current methods still struggle with balancing utility, access control, and reliable forgetting, highlighting the need for improved memory management in shared institutional applications.