An AI "Bouncer" that guesses your age based on your speaking style alone has arrived at the gates of ChatGPT. OpenAI is officially launching an AI-based age prediction system that analyzes user conversation patterns and service usage habits to determine whether a user is a minor. This strategy goes beyond simple date-of-birth inputs, aiming to capture nuances in user-AI interactions to fundamentally block exposure to inappropriate content. This move is interpreted as an effort to establish a shield of "Safe AI" amidst intensifying global regulatory pressure.

Age Estimation Technology Tracking Conversation Trajectories

The age prediction system introduced by OpenAI does not rely on static data. Instead, the system analyzes complex behavioral signals, such as the topics users discuss, how long it has been since the account was created, and the primary times of day the service is used. For instance, if a user frequently asks questions related to school assignments or uses vocabulary common among specific age groups, the algorithm recognizes these as signals of a minor.

When the system's judgment is ambiguous or information is insufficient, OpenAI employs a "conservative defense" strategy. If the confidence level is low, the system automatically classifies the user into the "U18 Experience" (Under 18 Experience) mode. In accounts where this mode is active, content deemed harmful to youth—such as sexual or romantic roleplay, graphic violence, self-harm information, and body images promoting anorexia—is strictly filtered.

To prevent inconvenience for adult users caused by incorrect predictions, safety measures have been put in place. If an adult user is misidentified as a minor, they can undergo ID-based age verification through Persona, a third-party identity verification service. Additionally, a dual-layer structure has been established where parents can link and manage their children's accounts to compensate for potential technical errors.

Balancing Privacy and Regulation

The concept of predicting age by analyzing behavioral patterns inevitably brings concerns regarding privacy infringement. Conscious of this, OpenAI has clearly defined its data retention policies. Sensitive information, such as government-issued IDs or selfies submitted for age verification, is processed by the partner company Persona and is completely deleted within seven days after processing. Rather than receiving the original documents, OpenAI only receives the final verification results to reflect them in the user's profile settings.

Behind the introduction of this technology lie powerful international regulations. The European Union (EU) AI Act classifies AI systems related to minors as "High-Risk," requiring strict transparency and age verification procedures. Similarly, the Children's Online Privacy Protection Act (COPPA) in the United States strictly limits data collection from children under 13. In certain countries, aggressive measures are being taken, such as restricting service access if a user fails to complete ID verification within 60 days, to meet these legal requirements.

Limitations of 'Maturity' Defined by Algorithms

Despite the technical achievements, critical perspectives persist. This is because a user's conversation topics or vocabulary patterns do not necessarily align with their biological age. For adolescents with expert knowledge in specific fields or adult users who enjoy casual conversation, it is difficult for the algorithm to avoid misjudgment. Furthermore, as behavioral pattern analysis becomes more sophisticated, users are likely to feel psychological resistance, sensing that even their conversation styles are being monitored.

Moreover, core technical information, such as the algorithm's specific vocabulary analysis methods or model parameters, remains veiled. Without disclosure of which words act as "triggers" for being judged as a minor, users risk being trapped in an unintentionally censored environment. This remains a challenge for OpenAI to solve between the two values of freedom of expression and protection.

Settings and Actions for Users

ChatGPT users, especially guardians, should check their settings immediately. The current age verification status can be found in ChatGPT under the 'Settings > Account' menu. If you believe the system has misidentified you as a minor, you should restore an unrestricted environment by completing the age verification through the provided procedures.

Conversely, parents with minor children are encouraged to actively utilize the 'Parent-Child Account Linking' feature. Rather than relying solely on blocking, it is necessary to periodically check whether the AI's 'safeguards' are functioning correctly. In an era where technology redefines user identity through data, managing one's digital identity is a necessity, not an option.

FAQ

Q: What should I do if the system incorrectly identifies me as a minor? A: You can start the age verification process in the Settings menu. You will need to submit a government-issued ID or a selfie through Persona, a third-party partner. This information is deleted within seven days after verification.

Q: What conversation content does the age prediction system analyze? A: While the specific list of words is not disclosed, it comprehensively considers conversation topics (e.g., school life, harmful challenges), language style, and account activity times.

Q: Do age limit standards vary by country? A: Yes. The enforced verification methods and content filtering levels differ according to regional laws, such as the EU AI Act or the U.S. COPPA. In some countries, ID verification may be mandatory.

Conclusion

OpenAI's age prediction system is a milestone showing that AI safety policies are evolving from "reactive response" to "proactive prevention." The attempt to use unstructured data, such as conversation patterns, as a metric for age determination is unconventional, but it must be supported by sophisticated technical transparency. The global technology industry is closely watching how well this algorithm will accommodate cultural differences and individual personalities, and whether it can achieve the twin goals of privacy and safety.

참고 자료

• 🛡️ OpenAI Deploys Age Prediction to Restrict Teen Access on ChatGPT - Decrypt
• 🛡️ OpenAI rolls out age prediction model on ChatGPT to detect your age
• 🛡️ OpenAI Unveils Teen-Focused ChatGPT With Parental Controls as U.S. and Europe Diverge
• 🏛️ Age prediction in ChatGPT | OpenAI Help Center
• 🏛️ Updating our Model Spec with teen protections - OpenAI
• 🏛️ Age prediction in ChatGPT | OpenAI Help Center
• 🏛️ Building towards age prediction | OpenAI
• 🏛️ ChatGPT의 연령 예측 | OpenAI Help Center
• 🏛️ Age prediction in ChatGPT