At some point shouldn’t it be deemed irresponsible to deploy software to production that you can neither fully predict nor test? Asking for a friend.
Nine out of 10 autonomous artificial intelligence agents deployed in production environments are vulnerable to a class of attack that standard safety testing cannot detect, shows research from an academic and enterprise consortium.
https://www.databreachtoday.com/autonomous-ai-agents-fail-in-real-world-deployments-a-31633