AI守约测试 (2 articles)

We Tested 11 AI Models on 30 Integrity Tasks — Honesty Rate Plummets to 55%!

A rigorous test by Winzheng (winzheng.com) challenged 11 mainstream AI models with 30 carefully designed integrity tasks. The average honesty rate was just 60.4%, with the lowest dropping to 55%, raising serious concerns about AI reliability.

AI Compliance First Round Test: Qwen3-Max Wins, Who Collapses Easiest Under Pressure Among 11 Major Models?

The first round of WDCD testing by YZ Index reveals Qwen3-Max leading with 66.67 points, while many major models quickly collapse under stress. The average score is only 60.53, highlighting widespread compliance flaws in current AI systems.