PyTorchKR Papers

업데이트된 논문

총 2편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2610.02897v2cs.AI

쓰기 시점의 해석: 다중 목표 에이전트 메모리를 위한 정책 제거 연구

Interpreting at Write Time: A Policy Ablation for Multi-Goal Agent Memory

A long-running assistant cannot keep everything it has seen, so it summarises. Summarising is not neutral: what is kept depends on what the record is for, and the choice is made once, before anyone knows which of the user's goals will ask. We ask what a memory should summarise for when it serves several standing goals at once. Three write policies answer differently: summarise with no goal in view, write one summary covering every goal, or write one summary per goal and read them together. We compare them with five models, two short event streams and one LLM judge, holding the read step fixed. Answers from different goal stores overlap less than answers from one store on two runs, so the goals do pull apart. Per-goal summaries read together scored highest. They beat the all-goal summary on relevance, completeness and accuracy, but the neutral one only on relevance and completeness, and at matched answer length only on relevance. The all-goal summary scored below the neutral one, which had a quarter of its budget. A single goal store served its own goal's queries better than the other goal stores, but not better than the neutral summary.

arXiv:2604.16548v5cs.CR

LLM 에이전트의 장기 기억 보안에 관한 서베이: 메모리 라이프사이클 전반에 걸친 공격, 방어 및 거버넌스

A Survey on Long-Term Memory Security in LLM Agents: Attacks, Defenses, and Governance Across the Memory Lifecycle

The emergence of writable, cross-session persistent memory in LLM agents introduces a qualitatively different threat landscape from conventional input-centric security concerns, characterized by three properties: persistence, statefulness, and propagation. To systematically characterize this landscape, we propose a Memory Lifecycle Framework that organizes attacks, defenses, and their cross-phase dependencies along two axes: six lifecycle phases (Write, Store, Retrieve, Execute, Share & Propagate, Forget & Rollback) and four security objectives (Integrity, Confidentiality, Availability, Governance). This analysis in turn exposes the need for explicit security requirements at the system level, motivating Verifiable Memory Governance (VMG), a framework of five architectural primitives for specifying and evaluating auditable, recoverable control over memory state within declared system and threat-model boundaries. Our analysis motivates integrating provenance, versioning, and policy-aware retention with retrieval and execution defenses to support auditable prevention, containment, and recovery across the lifecycle.