We handed two AI models Claude Fable 6 and Opus 4.8 the same 14 month-end close files, the same $1,035K cash variance, and the same prompt.
One found a material accounting anomaly on the first pass.
The other missed it entirely.
Same data, same question but different conclusions. And the difference was not cosmetic. It was the kind of difference that chang…
