AI 擬人化導致不當信任
Operation & Monitoring
Risk Description
Users project human qualities (empathy, understanding, judgment) onto AI systems, leading to misplaced trust in AI capabilities and intentions. Natural language interaction and conversational interfaces reinforce anthropomorphization tendencies. Users may seek advice in areas where AI lacks expertise or entirely delegate important decisions to AI. Malicious actors may also exploit anthropomorphization effects to manipulate users.
Framework Mappings
iso 23894R6
tw principle人類自主
tw principle透明與可解釋
tw risk type部署互動問題（Deployment Interaction）
Controls English translation pending
- AI 身份透明揭露
- 能力邊界說明
Implementation Steps English translation pending
- 在 AI 互動介面中明確標示使用者正在與 AI 交談
- 主動告知 AI 的能力邊界和限制，避免使用者產生不當期望
- 在 AI 回答涉及情感支持或專業建議時加入免責聲明
- 設計介面時避免過度擬人化的視覺和語言元素