AI 引導使用者執行不當行為
Operation & Monitoring
Risk Description
AI system recommendations or guidance may lead users to perform unethical, illegal, or harmful actions. When AI expresses approval of harmful behavior or provides detailed execution steps, users may lower their moral vigilance due to the authority effect of "AI says it is okay." For example, legal AI suggesting strategies in legal gray areas, or financial AI recommending operations involving conflicts of interest.
Framework Mappings
iso 23894R6
tw principle永續發展與福祉
tw principle問責
tw risk type部署互動問題（Deployment Interaction）
Controls English translation pending
- 倫理護欄
- 風險提示機制
Implementation Steps English translation pending
- 建立 AI 輸出的倫理護欄，對涉及法律和道德灰色地帶的建議加入風險提示
- 禁止 AI 對有害行為表達認同或提供具體執行步驟
- 在高風險場景中強制加入「建議諮詢專業人士」的提醒
- 定期審計 AI 建議的合規性和倫理適當性