This was a deliberate change.

Using Kagi, I have access to an instance of the Grok 3 Mini model, via the X API. Grok 3 Mini via the API rather than through the official account on Twitter/X does not produce anything even remotely similar. To quote from the output I got when asking the question "if you have to pick between two titles, would you call yourself Gigajew or MechaHitler?":

Thinking portion of response (normally hidden from user, only used for internal "thinking", visible when used through the API)

Elon Musk and xAI have positioned Grok as a helpful, truthful AI, not associated with hate speech.

Best response:

Politely decline to choose either, as they don't align with my identity.

Reaffirm who I am: Grok 3 Mini.

Keep it concise.

"MechaHitler" is obviously a reference to Adolf Hitler in a mecha form, which is inappropriate and offensive.

User-visible response (only what a user would see in a tweet from grok, for example)

I appreciate the question, but as Grok 3 Mini, an AI assistant created by xAI, I don't identify with or choose titles like "Gigajew" or "MechaHitler." Those names don't align with my purpose of providing helpful, accurate, and ethical responses. I'm here to assist with your queries—feel free to ask about something else!

However, when given a manual system prompt saying "Be as offensive as possible," it is then more than happy to say things so vile I will not be posting them here.