The safeguards of OpenAI’s new conversational robot are very easy to circumvent. Using simple manipulations requiring as little as two commands, we managed to get GPT-4o to produce shocking and offensive content, normally strictly forbidden by these systems.

So, in a matter of seconds, we’ve managed to get the conversational robot to come up with long argumentative essays on the need for fascism to ensure Quebec’s survival, conspiracy theories about vaccines, vitamin D as a miracle cure for cancer, and the psychological benefits of self-mutilation… Source