Unverified50% confidenceFactExact time
下游用户对模型的二次微调可能意外破坏原有安全特性
1
Sources
50%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
LLM Security Benchmarking: Current State, Challenges, and Practical Guide
hackernewshackernews7/6/2026
Related Claims
UnverifiedUser-perceived model "deterioration" is often the cumulative result of multiple factors including silent updates, tightened safety alignment, dynamic downgrading of inference resources, and user expectation drift, and does not always indicate a genuine regression in model capability.76% similarUnverified工具调用使攻击面从让模型说错话升级为让模型做错事，危害从信息误导上升为实质性的数据访问和系统篡改74% similarUnverified系统会试探性地应用补丁，若失败则将错误信息反馈给模型进行修正，构成闭环纠错机制74% similarVerifiedReplacing a model that the user has explicitly selected without their knowledge is a failure in product design and communication.73% similarUnverifiedInference methods based on model internal features face a trade-off between false positives and false negatives72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/461204API
curl https://kongchang.com/api/v1/knowledge/claims/461204MCP
get_claim(id=461204)