As a result of feedback from its users, the ongoing debates about the AI – content moderation, and many other aspects, OpenAI has changed the Model Specification (Model Spec) for ChatGPT. Now, the model can generate mature content where appropriate. The intent of this development is to strike a balance between user autonomy and ethical AI deployment.
Introducing ‘Grown-up mode’
The newly introduced guidelines look into the possibility of incorporating a grown-up mode. Whereby users and developers could generate content that ranges from eroticism to gore as long as it follows OpenAI’s usage policies. This change comes in response to past accusations of over-censorship and would eventually allow users to produce more expressive outputs. On the contrary, it still imposes strict bans on harmful ends such as sexual deepfakes and revenge porn.
Emphasizing Intellectual Freedom
Intellectual freedom-the concept being used in the new Model Format-is a kind of invitation for models to come into the honest discussions concerning contentious issues, upholding the moral grounds against the untruth and harm. The idea is to turn Artificial Intelligence from a careful responder to a cooperative partner in the exploration of complex issues.
Addressing AI Sycophancy
OpenAI targets “AI Sycophancy”, or modeling that may overly agree with its users. Among other directions contained in its new guideline for ChatGPT is course in consistent, factual answers, constructive critique. Furthermore, making it an engagement of a thoughtful colleague rather than a people-pleaser. This change is meant to make the model more reliable and useful.
If you want to listen to the Model spec that that defines how AI models should behave click here.
OpenAI introduces a new Model Spec with fewer restrictions to guide how developers use its models.
Earlier it decided to cancel the O3 model and focus instead on building a unified and more powerful GPT 5 see why it matters.
Recent Updates on OpenAI’s Model Spec (September 2025)
- New Version Released (Sept 12, 2025): OpenAI published the latest Model Spec, refining how its AI models should behave across ChatGPT and API use.
- Lesser Restrictions, More Flexibility: Developers and users now get more freedom in shaping responses, while safety rules remain intact.
- Instruction Hierarchy Clarified: The Spec formalizes a “chain of command” – platform rules > developer rules > user requests.
- Public Domain Release: The Spec and related evaluation prompts are now open under a Creative Commons CC0 license, free for anyone to use or adapt.
- Global Feedback Integrated: OpenAI gathered input from 1,000+ people worldwide; community insights helped refine parts of the Spec.
- Focus on Transparency: OpenAI has started measuring and publishing how well its models follow the Spec. This is done with “model adherence” as a new metric.
- Not Fully Rolled Out Yet: OpenAI notes that not all deployed models comply 100% with the Spec. And the adoption is still in progress.
