模型同質性引發系統性風險
Deployment
Risk Description
When multiple AI systems use the same or similar foundation models, they share the same biases, blind spots, and failure modes. In financial markets, multiple trading systems using the same model may simultaneously make identical decisions, triggering flash crashes. In healthcare, homogeneous diagnostic models may collectively misjudge the same rare cases. Model homogeneity amplifies individual system risks into systemic risks.
Framework Mappings
iso 23894R7
tw principle資安與安全
tw principle永續發展與福祉
tw risk type社會系統性影響（Societal Systemic）
Controls English translation pending
- 模型多樣化要求
- 系統性風險監控
Implementation Steps English translation pending
- 在關鍵基礎設施中要求使用多樣化的 AI 模型
- 建立跨組織的系統性風險監控機制
- 對行業內的模型集中度進行定期評估
- 制定模型依賴度壓力測試，模擬基礎模型故障的連鎖效應