OpenAI "Unbundles" ChatGPT Content Filter to Generate Adult NSFW Content Without Warnings

On Wednesday, February 15, OpenAI released the latest version of its "Model Guidance" document, which purports to fine-tune ChatGPT behavior/responses by "unbundling" content filters and allowing ChatGPT to skip content warnings (e.g., erotic and violent content) in certain situations or in "appropriate contexts" and directly generate adult NSFW content. The latest version of the "Model Guidance" document purports to fine-tune ChatGPT behavior/responses, "loosening" the content filters to allow ChatGPT to skip content warnings and directly generate adult NSFW content (e.g., erotic and violent content) under certain circumstances or in "appropriate contexts.
In fact, OpenAI has been secretly working on this feature since May 2024, when the model guidance document vaguely mentions that OpenAI is exploring "whether we can responsibly provide the ability to generate adult NSFW content in age-appropriate contexts, either through an API or directly on the ChatGPT web side! ".
Referring to the Reddit platform, some users' experiments have shown that ChatGPT has indeed loosened its content filters. OpenAI responded that people are still required to follow OpenAI's standard license, which prohibits adult NSFW content involving children.