can't people just point their AI to the latest pull request that includes the fix and ask it to deduce the vuln?
People could point their AI at a physics textbook and ask it to deduce the vuln from first principles too but that doesn’t mean it will be successful.
apparently not
Please do and report back here.