Primary sourcearXiv: OpenAI
View sourceEvent ·Official first-hand
ASIRF:免重训练的敏感信息脱敏智能体框架
arXiv 论文提出 ASIRF(Agentic Sensitive Information Redaction Framework),在推理时从知识库检索领域相关的敏感信息定义,无需重新训练即可适配新领域。论文实现了三次调用的多智能体流水线和单智能体两种架构,在 10 个小型开源模型、8 个数据集上评测,并以 OpenAI Privacy Filter(OPF)作为训练分类器基线。每个领域只需几十条专家编写的定义、零训练数据,ASIRF 在 80 个模型-领域组合中的 68 个(85%)召回率超过 OPF,落后场景主要集中于 OPF 的训练分布内领域。
Current read
脱敏工具换个领域就得重训练?这篇论文用智能体检索领域定义,几十条规则就能超越 OpenAI Privacy Filter 的 85% 组合。
11 sources53° AI heatLast updated 2026/9/24 08:02:48
Evidence
11 sourcesRelated source 1Xianbao Qian
View sourceRelated source 2The Information
View sourceRelated source 3Sam Altman
View sourceRelated source 4Gary Marcus
View sourceRelated source 5Tech in Asia
View sourceRelated source 6OpenAI Blog
View sourceRelated source 7Decoder
View sourceRelated source 8rohanpaul_ai
View sourceRelated source 9IT之家
View sourceRelated source 10kimmonismus
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.