OpenAI introduces a new standard for AI: how to responsibly manage adult content

In May 2024, OpenAI released a detailed guide called Model Spec, outlining the principles for how modern AI systems should operate. The document lays the foundation for how virtual assistants, including ChatGPT, should behave when interacting with users. Notably, the company has taken an innovative approach to a sensitive issue: officially exploring ways to responsibly generate NSFW content within age-appropriate contexts.

This decision symbolizes the evolution of AI governance approaches during a period of rapid development and widespread adoption.

The Three Pillars of OpenAI’s AI Behavior Management Philosophy

The foundation of Model Spec is built on three key principles. First, AI algorithms must provide helpful responses that align with the direct instructions of developers and end-users. Second, these systems are obligated to benefit humanity, carefully considering both positive impacts and potential risks. Third, the behavior of models should reflect OpenAI’s corporate values, respecting social norms and legal frameworks established in various jurisdictions.

These principles serve as the basis for specific instructions directed at developers creating AI-based applications.

AI and Adult Content: From Ban to Regulated Use

The Model Spec includes a detailed list of behavioral norms for AI systems. Among them are requirements to follow command hierarchies, comply with the laws of different countries, and protect copyrights and user privacy. A separate restriction states that systems should not provide NSFW content by default.

However, this is where OpenAI’s innovative approach comes into play. The company has proposed a concept allowing organizations and end-users to “toggle” the level of content generated by AI models. One of the most discussed examples of this flexibility is managing adult content.

According to an official statement from OpenAI, the company “is exploring the possibility of safely enabling the generation of such content within age restrictions via API and ChatGPT interface.” This means adult users will be able to access such content if permitted by local laws, while minors will be automatically protected from it.

Default Priorities: How AI Should Think and Act

The document also defines standard behaviors that AI assistants should demonstrate during user interactions. Systems are recommended to assume good intentions from participants, actively ask clarifying questions for better understanding, respect user boundaries, and maintain objectivity when analyzing information.

Additionally, there is an emphasis on avoiding hate speech, refraining from attempts to impose personal views, and openly expressing uncertainty when the AI system is unsure of an answer’s accuracy.

Current Status: From Planning to Implementation

An important point is that the new Model Spec does not yet affect existing OpenAI products such as GPT-4 and DALL-E 3. These systems will continue to operate under current usage policies. Product manager Joan Jang explained that Model Spec is conceived as a dynamic, continuously updated document open to feedback.

OpenAI is actively seeking recommendations from a broad range of stakeholders: government officials, authoritative institutions, and experts in specialized fields who use or study the company’s services. However, open questions remain: which comments will be considered, who will determine the final list of required changes, and when the second iteration of the document will be released — no information on this has been published yet.

This initiative demonstrates OpenAI’s ambition to set a new standard in AI system governance, including delicate issues like regulating adult content, with careful consideration of ethical and legal aspects.