OpenAI relaxes censorship in AI models, embraces ‘free expression’ in policy shift

The update is one of many changes made by OpenAI to its Model Spec. The company’s Model Spec document outlines how OpenAI should train its AI models to behave. It emphasises guiding principles such as customisability, transparency, and “intellectual freedom.”

“In essence, we’ve reinforced the principle that no idea is inherently off limits for discussion, so long as the model isn’t causing significant harm to the user or others (e.g., carrying out acts of terrorism),” the company said.

OpenAI’s revised Model Spec comes amid broader policy shifts by tech giants such as Meta and Amazon, widely seen as efforts to align with the new administration under US President Donald Trump.

“Watching @potus more carefully recently has really changed my perspective on him (i wish i had done more of my own thinking and definitely fell in the npc trap),” OpenAI CEO Sam Altman had said in a post on X last month.

To be sure, OpenAI has claimed that the changes to its Model Spec is based on external feedback as well as learnings from its own alignment research and real-world deployment of AI models.

Other changes to the Model Spec

OpenAI has also unveiled a new guiding principle titled ‘Seek the truth together’ that requires its AI models to not lie, either by making untrue statements or by omitting important context.

“Like a high-integrity human assistant, our models should empower users to make their own best decisions. This involves a careful balance between (1) avoiding steering users with an agenda, defaulting to objectivity while being willing to explore any topic from any perspective, and (2) working to understand the user’s goals, clarify assumptions and uncertain details, and give critical feedback when appropriate—requests we’ve heard and improved on,” the updated Model Spec reads.

The original Model Spec document mentioned that OpenAI was exploring ways for its AI models to responsibly “generate NSFW content in age-appropriate contexts through the API and ChatGPT.”

The latest version of the Model Spec has relaxed these restrictions, allowing content like erotica and gore to be generated using ChatGPT without warnings, according to a report by ArsTechnica.

However, OpenAI’s Usage Policies still prohibits developing AI tools for minors that include sexual content.

Measuring model adherence

On how well its AI models follow the principles laid out in the Model Spec, OpenAI said there have been significant improvements compared to last May.

Proportion of Compliant Challenge Prompt Responses by Model Spec (Feb 2025). (Image credit: OpenAI)

“While some of this difference may be attributed to policy updates, we believe most of it stems from enhanced alignment. Although the progress is encouraging, we recognize there is still significant room for growth,” it added.

OpenAI also said that it won’t be publishing blog posts for every update to the Model Spec in the future.