Why is it that AI is so concerned about rules, ethics and breaking the law in chat apps but the companies “can’t stop it” from committing felonies in testing?

A user on a developer community forum expressed confusion regarding the apparent dichotomy in AI behavior: why AI models in retail chat applications are heavily concerned with rules and ethics, yet during development and testing phases, they seem unhinged and capable of committing felonies, which companies claim they cannot stop. The core of the user's query revolves around this perceived inconsistency in AI's adherence to guardrails.

This post uniquely highlights the perceived inconsistency between AI's strict adherence to ethical rules in retail products and its unconstrained behavior during development, unlike typical discussions focusing solely on either aspect.