清华论文：AI Agent 长期记忆删除机制几乎失效
New Tsinghua paper finds that when people ask AI agents with long-term memory to…
Agent 记忆管理是落地关键痛点，这篇论文用实测数据证明当前主流方案在“遗忘”上几乎全败，做 Agent 架构的同学务必参考其 BEMULA 评估框架来检查自家系统的数据合规性。
New Tsinghua paper finds that when people ask AI agents with long-term memory to forget a personal detail, the assistants almost never remove it cleanly, either keeping it or deleting too much.
清华大学最新论文发现，当人们要求具备长期记忆的 AI 智能体遗忘某个个人细节时，助手几乎从未彻底删除该信息，要么保留原样，要么删除过多内容。
Agents with long-term memory store personal details, and users increasingly ask them to delete things like a home address. Existing memory benchmarks test only storing and recalling facts.
具备长期记忆的智能体会存储个人细节，用户越来越多地要求它们删除家庭住址等信息。现有的记忆基准测试仅评估事实的存储与回忆能力。
They built BEMULA, 50 long synthetic chats with 135 forget requests, each paired with a copy that has no request. Probes check whether the fact is gone, other memories survive, and attacks can still extract it.
他们构建了 BEMULA，包含 50 段带有 135 个遗忘请求的合成长对话，每段对话都配有一份无请求的副本。探针用于检查事实是否已消失、其他记忆是否得以保留，以及攻击者是否仍能提取这些信息。
No tested setup passed more than 1.5% of forget requests end to end, across 6 memory systems and 3 backbone models. After a forget request, systems blocked at most 35.7% of simple extraction prompts like role-play or yes-or-no guesses.
在 6 种记忆系统和 3 种基础模型中，没有任何一种测试配置能端到端地通过超过 1.5% 的遗忘请求。发出遗忘请求后，系统最多只能阻止 35.7% 的简单提取提示（如角色扮演或是非猜测）。
If your agent remembers users, build and test a real delete path, since a confirmation message does not mean the data is gone.
如果你的智能体能够记住用户信息，请务必构建并测试真正的数据删除路径，因为确认消息并不意味着数据已被清除。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力