OpenAI published Model Spec: How to control AI-generated porn and other explicit content

OpenAI’s Artificial Intelligence Laboratory has introduced the first version of its Model Spec document, which defines behavioral principles for AI models. The document includes detailed instructions on how to responsibly handle sensitive content, including issues related to AI-generated pornography and other explicit material. This marks a significant step for the company toward ethical AI development.

Key Principles for Managing AI Model Behavior

OpenAI has outlined three core principles guiding AI system operation. First, AI models should be developed to assist both developers and end-users in obtaining helpful responses according to their instructions. Second, algorithms should function in ways that benefit humanity while minimizing potential risks. Third, AI models should reflect OpenAI’s values and principles regarding social norms and applicable laws.

These three foundations form the basis for all other recommendations in the Model Spec. The company hopes that this structured approach will help create a more predictable and safe environment for working with AI technologies.

NSFW Content and Responsible Creation: OpenAI’s Position

One of the most discussed parts of the document is the section dedicated to managing explicit content. OpenAI does not outright ban working with AI-generated pornography but proposes a controlled approach. The company “is exploring the possibility of responsibly enabling the generation of such data in age-appropriate contexts via API and ChatGPT platform.”

The approach aims to allow companies and users to choose the level of “spiciness” of their AI models. This means developers can customize AI system parameters based on their needs and target audience. Product manager Joan Jung explained that the goal of the document is to gather public feedback on how models should behave and to draw a clear line between deliberate algorithmic actions and errors.

Five Key Rules for Developers

The Model Spec establishes a set of mandatory rules that all developers working with OpenAI’s AI systems must follow. Developers are required to adhere to instruction hierarchies, comply with current laws, and avoid creating informational threats. Additionally, respecting copyright and intellectual property, protecting user privacy, and not generating NSFW content without proper restrictions and permissions are essential.

These rules apply both to working with AI-generated pornography and other sensitive content categories. The document also recommends that models default to assuming users’ good intentions, ask clarifying questions when needed, avoid crossing established boundaries, maintain objectivity, and express uncertainty when appropriate.

Impact on Existing AI Models

It is important to note that, at this time, the Model Spec will not affect already released OpenAI products such as GPT-4 and DALL-E 3. They will continue to operate under existing usage policies. However, the document is intended as a “living” tool that the company plans to update frequently based on feedback from the public, policymakers, academic institutions, and experts across various fields.

The company expects to gather opinions from all interested parties, including those using OpenAI’s services. It is currently unknown which feedback will be considered and who will determine necessary changes to the document. There is also no information yet about the release of a second version of the Model Spec.

Future Outlook: AI Porn and New Standards

The Model Spec represents an important step toward standardizing approaches to managing AI system behavior. The document demonstrates that OpenAI takes ethics and safety seriously in AI development, including delicate issues related to AI-generated pornography and explicit content. Instead of a complete ban, the company has chosen a path of responsible management, allowing developers to make informed decisions.

This approach reflects a growing industry understanding that full control over content is impossible and often counterproductive. Instead, more effective strategies are based on transparency, accountability, and stakeholder engagement. As AI becomes an increasingly widespread tool, documents like the Model Spec will play an ever more important role in shaping standards and practices for handling AI-generated pornography and other sensitive content worldwide.