which kernel devs? I only saw Jim Zemlin and other executives quoted.
I suspect they aren't releasing it due to inference costs. This is spin.
If this was true at face value, Claude code's own source would look _very_ different. I suspect the false positives are underreported. I bet they send a filtered set to cybergym. ARR is proof enough they cook metrics.
With the Firefox vulnerabilities, no false positive rate was given iirc. All 3 disclosures needed the sandbox disabled.