[Bug/建议] GLM API 1301 内容护栏误伤自动化提示词——领导人姓名出现在合规工作流指令中导致整会话拒绝
Nobody has claimed this yet.
- Dominant language
- No language data
- Stars
- 22
- Forks
- 1
- PR merge metrics
- No merged PRs in 30d
Description
现象
ZCode 定时自动化(每日政策扫描,合法投研工作流)今晚 00:15(北京)触发即死,报错:
[1301][系统检测到输入或生成内容可能包含不安全或敏感内容](Trace: 2026090500202657dce70195f34fdd)
根因
自动化提示词的信号分级规则里含具体领导人姓名(如"署名习近平→最高权重",源自政策扫描方法论:元首级署名政策=最高信号)。同晚会话的其余 4 个自动化(提示词无该姓名)全部成功——唯一差异变量就是这个名字,可定位为护栏在输入层对指令文本中领导人姓名的硬拦截。
影响
- 中文政经类投研/舆情/政策研究工作流无法完全回避这类姓名(新闻原文自带);当前行为=整条请求拒绝→自动化会话静默死亡,产出缺失。
- 拒绝发生在会话中途时用户只看到一行报错,无法定位是哪段内容触发。
建议
- 对"指令/规则文本中出现领导人姓名"这类非攻击性用法放宽或改为人工审核队列——新闻机构/金融研究场景姓名是常态;
- 1301 报错信息中附带触发内容摘要或位置(哪一段/哪个工具结果),否则自动化无法自愈;
- 提供会话级降级策略选项:1301 时跳过触发块继续,而非整会话终止。
临时 workaround(已实施,供其他用户参考)
配置/提示词一律用通称("元首/总书记"),人名识别交给新闻原文——语义无损,护栏可过。
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reproducing the 1301 failure with the quoted automated prompt and compare it with the four successful automations; confirm whether the leader's name is the only differing input. The issue proposes allowing non-attack use, adding trigger-location details, or continuing after a blocked block, but names no files, tests, or implementation entry point, so the definition of done is unresolved.
Written by the indexing model from the issue text.
Assessment
- Domain
- api, security
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100