Cascading Amplification of Errors
串聯錯誤逐層放大
Operation & Monitoring
Risk Description
When a front-end system misreads a piece of information, and all subsequent analysis and decisions build upon it; due to unmitigated control gaps, the final report is well argued with clear source attribution, leaving reviewers no way to detect that the underlying premise was wrong, triggering compliance exposure and operational reputational costs.
Framework Mappings
MITRE ATLASAML.T0051
NIST AI RMFMEASURE 2.6
ISO/IEC 23894§6.4
OWASP Top 10 for LLMLLM06
MAESTRO
ISO/IEC 5338運作與監控
MIT AI Risk RepositoryDomain 7
Risk Treatment & Implementation Guidance
Verify cross-system outputs so downstream systems check upstream input credibility rather than fully trusting it; Track confidence propagation so front-end uncertainty carries through to final output; Set independent verification points for premises in critical chains