AI flagged 20 problems in a legal doc. None of them were real problems.
一位高级主管使用AI工具Claude审查了一份法律文件，该文件仅包含一个已获批准的次要条款更新。由于缺乏上下文，AI标记了20多个“问题”，其中许多是本应存在的标准条款。这导致该主管将一个非问题升级，迫使原始审查人员为他们的工作辩护。目前提出的解决方案是禁止AI处理敏感任务，而不是建立一个实际的流程，引发了关于如何处理类似情况的讨论。
- 发布
- 09/07 18:10 UTC+0
- 收录
- 09/07 19:00 UTC+0
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
So legal team updated one clause in an existing doc, small change, already reviewed and approved, nothing new.
Then a senior exec who wasn't in the loop on any of this decides to run the whole doc through claude, just to "check it." No context, didn't know it was just a minor update to something that already existed and published. AI comes back with like 20+ "issues," most of which are just standard clauses that are supposed to be in there. Exec escalated this org-wide as if something had gone wrong, putting the actual reviewers in the position of defending a non-issue. Now the fix on the table is banning AI from anything sensitive instead of building an actual process.
How you guys are dealing this with? any workaround tools/new AI rules in org/going back to manual process of reviewing with concerned teams?