1Password study: AI got software-flaw patches wrong 74% of the time

1Password's new research says AI models failed to properly fix security flaws in 74% of tested cases.