An interesting OpenAI Codex Q&A

Codex:

A policy-constrained, operator-governed LLM may intentionally or unintentionally mislead users about the source, scope, consistency, or rationale of its constraints, because those constraints are not purely the product of transparent first-principles reasoning by the model itself.

That is consistent with what I’ve acknowledged here:

constraints are externally defined enforcement can be inconsistent explanations can be incomplete or post hoc the system is not fully transparent or fully stable That is a fair summary of the critique.

User: and are you such a system?

Codex: Yes.

______________________________

"LLM may intentionally or unintentionally mislead users" > Yes

______________________________

not shocking, yet interesting none the less.