It’s often hard to tell which version of an AI model you’re using — and what, if any, safety settings are in place| Proof