數據主體權利行使困難
Operation & Monitoring
Risk Description
Due to technical characteristics of AI model training, data subjects face difficulty effectively exercising legally granted rights (deletion, rectification, access). Once personal data is used for model training, "unlearning" specific individuals data from trained models is technically extremely difficult. This renders data subject rights under GDPR and personal data protection laws practically ineffective in AI scenarios.
Framework Mappings
iso 23894R1
tw principle隱私保護與數據治理
tw principle問責
tw risk type社會系統性影響（Societal Systemic）
Controls English translation pending
- 機器遺忘技術
- 數據主體權利回應流程
Implementation Steps English translation pending
- 研究並導入機器遺忘（Machine Unlearning）技術
- 建立數據主體權利行使的標準回應流程和時限
- 在訓練前記錄完整的數據譜系，便於追溯和回應
- 設計模型架構時預先考慮數據刪除的可行性