OpenAI Implements Age Prediction Model in ChatGPT to Enforce Safety Restrictions

OpenAI has begun rolling out an age prediction model within ChatGPT designed to automatically detect users' ages and enforce content restrictions for minors. This system analyzes conversation patterns and usage timing to identify potential under-18 users, triggering protective filters without requiring upfront age verification.

The AI-driven detection examines factors including conversation topics initiated by users and the times of day when ChatGPT is accessed. Based on behavioral patterns, the model assigns users to either adult or teen categories. OpenAI acknowledges potential inaccuracies, noting that adults exhibiting teen-like behavior patterns might be incorrectly classified and subjected to restrictions.

When the system flags an account as underage, ChatGPT automatically implements safeguards that block access to content categories including:

• Graphic depictions of violence or gore
• Viral challenges encouraging dangerous behavior
• Material promoting extreme beauty standards
• Content advocating unhealthy dieting or body shaming

Adults mistakenly classified as minors can undergo verification at chatgpt.com/verify_age. The process involves:

1. Submitting a live selfie via webcam or mobile device
2. Uploading a government-issued ID (driver's license, passport, or state ID)
3. Automated matching of selfie to ID photo by OpenAI's verification partner, Persona

Upon successful verification, ChatGPT removes safety restrictions within minutes. Persona deletes all verification data within seven days, retaining only the confirmation of adulthood. OpenAI emphasizes that verified users retain full access to ChatGPT's educational and creative tools regardless of age classification.

The rollout impacts all ChatGPT accounts globally, with the system applying automatically. While the model reduces exposure to harmful content for minors, OpenAI recommends parental supervision for young users and reminds adults to verify their age if incorrectly flagged by the system.