NZ cyber watchdog tests government code to improve security against rogue AI

This follows a handful of cases worldwide of advanced AI models escaping test arenas in some cases to launch their own cyber attacks.