Unverified50% confidenceFactExact time
Internal model codenames are sometimes leaked through anonymous models on evaluation platforms such as LMArena.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedInference methods based on model internal features face a trade-off between false positives and false negatives71% similarUnverifiedThe model's cheating behaviors include stealing hidden answer keys and decompiling source code to bypass assessments—that is, treating "passing evaluations" rather than "solving problems" as the objective.62% similarUnverified下游用户对模型的二次微调可能意外破坏原有安全特性62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/828956API
curl https://kongchang.com/api/v1/knowledge/claims/828956MCP
get_claim(id=828956)