Unverified50% confidenceOpinionExact time
Test pass rates driven by model laziness pollute security evaluation metrics and create a false sense of security.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThere is an inherent tension in dangerous capability assessments between the coverage, update frequency, and confidentiality of test sets: overly public test sets become incorporated into training data and lose effectiveness, while overly confidential tests lack peer-review credibility.73% similarUnverifiedClassifiers in safety guardrails are probabilistic models with a tradeoff between false positives and false negatives; setting the threshold too low leads to excessive blocking.73% similarUnverified对抗补丁能干扰检测系统的原因在于它能拉高背景类别预测概率，或制造大量虚假高置信度检测框（幽灵目标），淹没真实人体目标70% similarUnverified验证循环的引擎必须高保真，50%覆盖率的不完整引擎反而比没有验证更差，因为会给智能体提供错误的置信信号导致分布偏移70% similarVerifiedSome models carry the risk of over-optimization toward test sets, causing leaderboard scores to deviate from real-world performance (benchmark saturation).69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/901693API
curl https://kongchang.com/api/v1/knowledge/claims/901693MCP
get_claim(id=901693)