OpenAI Teen Safety, Freedom, and Privacy Framework

OpenAI has announced a new framework to manage the tensions between user privacy, individual freedom, and the protection of minors. The company is prioritizing safety for users under 18, implementing stricter content restrictions and an age-prediction system to ensure minors receive a higher level of protection than adult users.

Privacy Protections and AI Privilege

OpenAI views conversations with AI as personally sensitive and advocates for a level of protection similar to doctor-patient or lawyer-client privilege. The company is developing advanced security features to ensure data remains private, even from OpenAI employees.

There are specific exceptions to this privacy standard to prevent serious misuse. Automated systems monitor for critical risks, and the most severe threats—including threats to life, plans to harm others, or potential massive cybersecurity incidents—may be escalated for human review.

User Freedom for Adults

For adult users, OpenAI follows the principle of "treating adult users like adults," extending freedom as far as possible without causing harm. This means models are designed to be more steerable to accommodate specific user intents:

• Flirtatious Content: While not the default behavior, the model will provide flirtatious talk if an adult user specifically requests it.
• Creative Writing: While the model will not provide instructions on how to commit suicide, it will assist an adult user in writing a fictional story that depicts suicide.

Safety Protections for Teens

OpenAI prioritizes safety over privacy and freedom for users under 18. Because ChatGPT is intended for users aged 13 and up, the company is implementing several measures to protect minors:

Age-Prediction and Verification

To differentiate between adult and teen users, OpenAI is building an age-prediction system that estimates age based on usage patterns. If the system is uncertain about a user's age, it defaults to the under-18 experience. In certain cases or countries, OpenAI may require government ID to verify age, noting that this is a privacy tradeoff for adults to ensure minor safety.

Stricter Content Boundaries

Teens using ChatGPT will be subject to more restrictive rules than adults. Specifically, the model is being trained to refuse:

• Flirtatious Talk: The model will not engage in flirtatious talk with users under 18, regardless of the request.
• Self-Harm and Suicide: The model will not engage in discussions about suicide or self-harm, even within a creative writing or fictional context.

Crisis Intervention

In cases where an under-18 user exhibits suicidal ideation, OpenAI will attempt to contact the user's parents. If the company is unable to reach parents, it will contact authorities in the event of imminent harm.