wuyang
1.6 KiB
1.6 KiB
Paper: MemLineage: Lineage-Guided Enforcement for LLM Agent Memory
type: paper title: "MemLineage: Lineage-Guided Enforcement for LLM Agent Memory" authors: Ciyan Ouyang, Rui Hou year: 2026 venue: arXiv url: https://arxiv.org/abs/2605.14421 code_url: source: arxiv collected_at: 2026-07-08 published_at: 2026-05-14 updated_at: 2026-05-14 status: skimmed relevance: high topics:
- agent-evaluation
- computer-use
- memory
- tool-use methods:
- cryptographic-provenance
- derivation-lineage
- sensitive-action-gate benchmarks:
- custom-cross-session-memory-harness
- AgentDojo-bridge models:
datasets:
- cs.CR
- cs.AI related_concepts:
- memory-provenance
- memory-poisoning related_jobs:
related_experiments:
- KC-001-agent-memory-pilot related_projects:
- learning/agent-memory collection_score: 17 collection_queries: agent-memory
One-line Takeaway
签名只能证明谁写入；敏感行动还必须检查当前理由是否沿派生链继承了不可信来源。
Pilot Skim
- problem: 恶意外部内容可以经多次 memory derivation 被语义洗白，未来再触发敏感动作。
- method: 为条目附加签名与 derivation DAG，把来源 trust 传播到 action gate。
- evidence: 自定义 deterministic harness 的三类攻击中，只有 lineage 方案全部达到 0 ASR。
- boundary: headline 是机制隔离 benchmark，不是完整公共 benchmark；未完成大规模 adaptive attack，且假设模型推理路径可信。