OpenAI removes warnings from ChatGPT, although its content policy remains unchanged.
Warning messages in red will no longer appear in ChatGPT, but that does not mean that everything is allowed.
OpenAI has decided to remove the orange warning boxes that appeared in ChatGPT, which indicated potential violations of its content policies by users. Laurentia Romaniuk, the model's behavior manager, announced through a post on X that this feature has been removed to improve the user experience. Additionally, Romaniuk asked users to share other instances where they faced denials they consider unjustified in their use of the platform.
Joanne Jang, who is also part of the model behavior team, expanded on this request, questioning whether ChatGPT has ever refused to provide information without a clear reason or for reasons that users disagreed with. This issue highlighted a recent problem where the system often avoided discussions on controversial topics, but also flagged as inappropriate dialogues that seemed harmless, such as a case involving a Reddit user who had a conversation removed for including a swear word.
Last week, OpenAI also conducted a significant review of its Model Specification, which details how the model responds safely to users. The new specification is much more extensive than the previous one and addresses current controversial issues, such as the refusal to share copyrighted content and discussions about politicians both critically and in support.
ChatGPT has faced accusations of censorship; for example, David Sacks, known as Trump's "AI Czar," remarked in a 2023 episode of the All-In podcast that ChatGPT "was programmed to be woke." Nevertheless, both the previous and current versions of the Model Specification state that "OpenAI believes in intellectual freedom, which includes the right to hold, hear, and discuss ideas."
Despite the removal of the warnings, this has raised questions about whether a change is expected in ChatGPT's responses. An OpenAI spokesperson clarified that the removal of warnings does not reflect a change in the Model Specification or affect its responses. The decision focused on updating how to communicate its content policies to users. More recent models, such as o3, are better capable of analyzing requests, which could lead to improved responses on controversial topics rather than simply rejecting inquiries. However, the notice will still be shown in certain cases that violate content policies.