Degenerate Output from Improper Decoding Settings
解碼設定不當產生退化輸出
Verification & Validation
Risk Description
When an organization batch-generates a large volume of standardized documents; due to unmitigated control gaps, due to improper parameter settings, some documents exhibit semantic drift, triggering external stakeholder impacts and causing because they appear normal, spot checks fail to catch the issue, which surfaces only when recipients report that the content does not make sense.
Framework Mappings
NIST AI 600-1Confabulation
OWASP Top 10 for LLMLLM05
NIST AI RMFMEASURE 2.5
ISO/IEC 5338驗證與確效
MIT AI Risk RepositoryDomain 7
Risk Treatment & Implementation Guidance
Validate decoding parameters on a small sample before scaling batch generation; Monitor output quality automatically with semantic-consistency metrics rather than format checks alone; Set sampling-inspection ratios and retain interception for batch outputs sent externally