Memento：不改权重即可恢复被模型编辑抹去的知识
The Truth Beneath the Tokens: Prompt-Space Recovery of Knowledge Edits in Language Models
做了模型编辑以为知识删干净了？这篇论文证明一句提示词就能恢复八成被编辑的事实。做安全审计或部署“已删除知识”模型的团队都得看看这个方法。
论文提出 Memento，一种纯输入空间的知识恢复机制，用于审计经过模型编辑的 LLM。它学习一个可复用的恢复信号，将编辑后的模型引向编辑前的回答，既支持连续恢复前缀，也支持可直接用于文本接口的离散文本提示，无需修改或分解模型权重。
在 3 个 LLM 和 5 种编辑技术上的实验显示，Memento 能恢复 80.07% 的留出编辑事实，相比基线有 9.58% 的相对提升，说明编辑可能只是压制了事实而未真正消除其可恢复性。此外，在未编辑的原始 LLM 上，Memento 相比标准推理将平均事实召回提升 53.9%。
这项工作对 AI 安全审计有直接意义：仅靠输入端干预就能唤醒被删除的知识，意味着基于模型编辑的“知识删除”可能无法彻底清除敏感信息。作者承诺开源源代码。
The Truth Beneath the Tokens: Prompt-Space Recovery of Knowledge Edits in Language Models
Model editing provides an efficient way to update factual knowledge in transformer models. Recent approaches reverse edits by modifying, masking, or decomposing model weights, but make recovery dependent on weight-space interventions. We instead show that pre-edit knowledge can be recovered through input-space interventions alone, without modifying the edited model's weights. We introduce Memento, a prompt-space mechanism for auditing residual knowledge that learns a reusable recovery signal to steer an edited model toward its pre-edit response. Memento supports both continuous recovery prefixes and discrete textual prompts for standard text interfaces. Across three LLMs and five editing techniques, Memento recovers 80.07% of held-out edited facts, achieving a 9.58% relative improvement over baselines, suggesting that editing may suppress the facts without fully eliminating their recoverability. We further investigate whether prompt-space recovery extends beyond model editing. On pristine, unedited LLMs, Memento improves average factual recall by 53.9% relative to standard inference. These results demonstrate the potential of prompt-space recovery for auditing residual knowledge in edited models and investigating the gap between factual knowledge and its accessibility during inference. We will release our source code.
本条情报归属 AITOP「论文」核心栏目，重点聚焦北美前沿实验室与硅谷科技巨头。在跨区域产业观察坐标中，该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」「各国 AI 监管法案、合规治理与主权算力政策」「中美大模型技术代差与商业化对照」研究主线，反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括：Memento、模型编辑、AI安全、LLM，读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明：本文原始线索采编自「arXiv cs.AI」（@Manit Baser, Rachel Yeo Hui Min, Aditya Nawal, Dinil Mon Divakaran, Mohan Gurusamy），经 AITOP 多维评分管线结构化提炼。
