OpenAI Deploys Age Prediction Model: ChatGPT Now Identifies Minors Through Behavioral Analysis
OpenAI is deploying an AI-powered age prediction model across ChatGPT to automatically identify and protect minors, moving beyond the “honor system” that previously relied on users’ self-reported birthdates. The system analyzes behavioral signals to trigger content restrictions for users it identifies as under 18—a critical shift given that Pew Research Center data shows 64% of U.S. teens now use AI chatbots, with 28% using them daily.
How AI Age Detection Technology Works
OpenAI’s age prediction system represents a departure from traditional age-gating methods like date-of-birth fields or credit card checks. Instead of relying on a single verification event, the model continuously processes behavioral and account-level signals to build a probabilistic age estimate. The approach draws on established research in behavioral biometrics, where machine learning models analyze interaction patterns that differ systematically between age groups.
The system evaluates multiple data streams simultaneously:
- Account longevity: How long the account has existed and its registration context
- Activity timing: Typical times of day when the user is active, which correlate with school schedules and parental oversight patterns
- Session behavior: Length, frequency, and cadence of interactions across the platform
- Usage patterns: Behavioral trends observed over time, including the types of queries submitted
- Stated age: The birthdate provided during registration, weighted against behavioral evidence
Academic research supports the viability of this approach. A cross-sectional study published in the Journal of Medical Internet Research found that AI models trained on smartphone keystroke dynamics could estimate user age with a mean absolute error of just 3.69 years using long short-term memory (LSTM) neural networks. Younger users consistently typed faster and more frequently, while older users showed slower, more variable typing patterns.
Beyond typing speed, behavioral age estimation can incorporate vocabulary complexity, sentence structure, topic selection, and even emoji usage frequency. Adolescent users tend to engage with shorter query formats, use more informal language, and demonstrate different topical interests than adult users. OpenAI has not disclosed the specific weighting of these signals, but the company has confirmed the model processes “behavioral and account-level signals” rather than content analysis alone.
How Age Prediction Works in Practice
Rather than trusting the age users provide during sign-up, OpenAI’s model examines the full behavioral fingerprint of an account. The system operates continuously—not as a one-time gate but as an ongoing assessment that can reclassify users as new data accumulates.
When the algorithm determines a user is likely under 18, ChatGPT automatically applies content filters—building on ChatGPT Atlas agent mode capabilities—restricting discussions of sex, violence, self-harm, and other sensitive material, regardless of what age the user claimed at sign-up. OpenAI has stated the system is deliberately calibrated to tolerate false positives over false negatives: flagging an adult as a minor is considered preferable to exposing a child to harmful content.
Notably, OpenAI has not published accuracy benchmarks, false positive rates, or any quantitative performance metrics for the system. As The Register noted, “We don’t know if the system correctly identifies 90% of users or 50%.” This lack of transparency has drawn criticism from researchers and advocacy groups alike.
Protections Applied to Minors
Users identified as minors receive an age-appropriate ChatGPT experience with specific safeguards that OpenAI detailed in its updated Model Spec:
| Protection | Description |
|---|---|
| Content filtering | Blocks graphic sexual content and violence |
| Roleplay restrictions | Prevents romantic roleplay and first-person intimacy |
| Sensitive topic handling | Extra caution with body image and eating behaviors |
| Safety prioritization | Protection over user autonomy when risks appear |
| Caregiver transparency | Avoids advice helping teens hide risky behavior |
| Parental controls | Parents can set quiet hours, control memory and training features |
| Crisis detection | Notifications to parents if signs of acute distress are detected |
In cases of acute distress, the system may involve crisis resources or, in rare situations, law enforcement to ensure safety. Parents who link their accounts can customize their teen’s experience, including setting time-of-day restrictions when ChatGPT cannot be used.
Verification for Misidentified Adults
Adults incorrectly flagged as minors can restore full access through identity verification with Persona, a third-party verification service. The process offers two options:
- Selfie verification: The default method using facial analysis to estimate age
- Government ID: Available if selfie verification fails or the user prefers document-based proof
OpenAI emphasizes that Persona deletes verification data within hours and never shares selfies or IDs with OpenAI directly—the company receives only a date of birth or age estimate. However, the requirement to submit biometric data or government identification to a third party has itself become a point of contention among privacy advocates.
Why OpenAI Is Acting Now
The age prediction rollout comes amid mounting regulatory and legal pressure on AI companies to protect young users. The urgency is underscored by usage data: according to Pew Research’s February 2026 survey, 72% of U.S. teens aged 13-17 have interacted with AI chatbots, with 12% reporting they use them for emotional support—a use case that raises particular safety concerns.
| Development | Impact |
|---|---|
| FTC investigation | September 2025 inquiry into AI chatbots’ effects on children targeting OpenAI, Meta, Alphabet, xAI, Snap, and Character.AI |
| Wrongful death lawsuits | Multiple families suing AI companies alleging chatbots contributed to teen suicides |
| Content generation bugs | April incident where ChatGPT generated inappropriate content for minors |
| COPPA Rule amendments | FTC finalized major COPPA updates in April 2025, effective June 2025, with compliance deadline April 2026 |
| FTC age verification workshop | January 28, 2026 workshop bringing stakeholders together on age verification tools and COPPA interplay |
| Planned “adult mode” | OpenAI preparing NSFW content features requiring robust age-gating first |
The FTC specifically requested information about how companies monetize user engagement, generate outputs, and mitigate negative impacts on young people. In February 2026, the FTC issued a policy statement announcing it will not bring enforcement actions against operators that collect personal information solely for age verification purposes—provided they do not repurpose, retain unnecessarily, or insecurely disclose that data.
Updated COPPA Rules Reshape the Landscape
The FTC’s amended COPPA Rule, finalized in April 2025 and enforceable by April 22, 2026, introduces several provisions directly relevant to AI platforms:
- Expanded personal information definition: Now includes biometric identifiers (voiceprints, facial templates), persistent identifiers, and behavioral or inferred data
- Separate consent for AI training: Disclosures of children’s data to train or develop AI technologies require separate, verifiable parental consent
- Third-party disclosure separation: Consent for advertising-related data sharing must be obtained separately from consent for primary platform functions
- Data minimization requirements: Operators must limit collection to what is reasonably necessary for the child’s requested activity
For OpenAI, these rules create a compliance imperative. Without reliable age detection, the company faces potential liability for collecting minors’ conversational data—data that could constitute “behavioral or inferred data” under the expanded COPPA definition—without proper parental consent.
California’s Parents & Kids Safe AI Act
On January 9, 2026, OpenAI and Common Sense Media announced joint support for the Parents & Kids Safe AI Act, described as the most comprehensive youth AI safety measure in the United States. The California ballot initiative would require:
- Age assurance technology with automatic protections for predicted minors
- Independent safety audits reported to the California Attorney General
- Ban on child-targeted advertising
- Prohibition on selling or sharing children’s data without parental consent
- Restrictions on AI companion chatbots suggesting isolation from family
- Prevention of AI systems claiming sentience or engaging in simulated romantic interactions with minors
The measure requires 546,651 signatures by June 25, 2026, to qualify for the ballot. OpenAI’s public endorsement signals the company’s strategy to shape regulation proactively rather than react to it.
Global Regulatory Landscape
OpenAI’s age prediction deployment exists within a rapidly expanding global framework of age verification requirements. The following table summarizes the current state of regulation across major markets as of early 2026:
| Jurisdiction | Key Legislation | Age Threshold | Scope | Status |
|---|---|---|---|---|
| United States (Federal) | COPPA Rule (amended 2025) | Under 13 | All commercial websites/apps collecting children’s data | Compliance deadline April 2026 |
| United States (States) | 25+ state laws (Virginia, Texas, Nebraska, etc.) | Under 16-18 (varies) | Social media and adult content platforms | Half of U.S. states now mandate age verification |
| European Union | Digital Services Act (DSA) + national laws | Under 16 (varies by member state) | Very large online platforms; adult content in France, Germany | Active enforcement; EU Digital Identity Wallet beta 2025, full rollout 2026 |
| United Kingdom | Online Safety Act (OSA) | Under 18 | Platforms hosting harmful or adult content | Enforced since July 25, 2025; accepts photo ID, facial estimation, bank/mobile checks |
| Australia | Online Safety (Social Media Minimum Age) Act | Under 16 | Designated social media platforms | Effective December 10, 2025; stricter verification from March 9, 2026 |
| France | National age verification law | Under 18 | Adult content sites; social media legislation planned | Effective June 2025; social media bill targeting September 2026 |
This patchwork of regulations creates compliance challenges for global platforms. The EU’s delayed deployment of OpenAI’s age prediction—announced to follow the global rollout by several weeks—reflects the additional requirements imposed by GDPR and the DSA around automated decision-making and biometric data processing.
Privacy Concerns and Civil Liberties Criticism
Digital rights advocates have raised substantial concerns about the approach. The criticism extends beyond OpenAI’s specific implementation to the broader normalization of behavioral surveillance as a child safety mechanism—a trend that accelerated sharply in 2025, which the Electronic Frontier Foundation called “the year states chose surveillance over safety.”
Alexis Hancock, director of engineering at the Electronic Frontier Foundation, stated:
“OpenAI is taking the moment to further train an age prediction model, where a false prediction will fall on the user to give private information to further verify their age to another company.”
— Alexis Hancock, Electronic Frontier Foundation
Hancock noted that the model “is not obligated to be correct, nor can the decisions be challenged,” raising questions about accountability when predictions are wrong. This lack of an appeals mechanism—beyond submitting biometric data to Persona—means users have no way to contest the algorithm’s judgment without surrendering additional personal information.
The ACLU has argued that age verification requirements burden the First Amendment rights of adults while doing little to protect children. The organization points out that determined minors routinely circumvent age gates—as demonstrated by Roblox’s struggling age verification system—while adults face privacy risks from submitting government IDs and biometric data to third-party services.
Mozilla similarly cautioned that “fundamental tensions around effectiveness, accessibility, privacy, and security have not been resolved” in age verification technologies. Specific concerns include:
- Surveillance normalization: Continuous behavioral monitoring sets a precedent that extends beyond child safety into general user profiling
- Data breach exposure: Verification records linked to personal identity create high-value targets for attackers
- Chilling effects: Users may self-censor queries on sensitive topics—health, sexuality, legal questions—knowing their behavior is being analyzed
- Disproportionate impact: Users without government ID or those uncomfortable with biometric submission have no recourse if misclassified
- Function creep: Behavioral profiling data collected for age estimation could be repurposed for advertising segmentation or content personalization
The June 2025 U.S. Supreme Court decision in Free Speech Coalition v. Paxton upheld Texas’s age verification law, setting a precedent that the EFF described as trampling free speech and undermining privacy. That ruling has emboldened state legislatures: half of U.S. states now mandate some form of age verification for online platforms, up from a handful in 2023.
Rollout Timeline and Future Plans
The age prediction model is rolling out globally now, with European Union deployment following in the coming weeks to meet regional requirements under the DSA and GDPR. OpenAI plans to refine prediction accuracy over time using rollout data—a process that itself raises questions about whether user interactions are being used as training data for the age estimation model.
This implementation clears the path for OpenAI’s “adult mode,” which CEO of Applications Fidji Simo indicated would debut in Q1 2026. That feature will permit users to create and access NSFW content—making robust age-gating a prerequisite. The commercial logic is clear: OpenAI needs age verification infrastructure before it can safely monetize adult content without regulatory exposure.
What This Means for Marketers
With ChatGPT’s expanding capabilities and the recent announcement of ads in the free tier, age prediction has direct implications for advertisers navigating an increasingly regulated landscape:
- Audience segmentation: Ads may only reach verified adult users, affecting targeting options and reducing total addressable audience by an estimated 15-20% based on teen usage rates
- Brand safety: Advertisers gain assurance their messages will not reach minors in inappropriate contexts—a significant concern given that brand safety standards are tightening across platforms
- COPPA compliance: The amended COPPA Rule’s April 2026 deadline means advertisers must ensure their AI platform partners have adequate age-gating before running campaigns that collect behavioral data
- Platform credibility: Robust safety measures may accelerate brand adoption of ChatGPT advertising, particularly among regulated industries like finance and healthcare
- Cross-platform implications: As competitors like Meta, Google, and Anthropic develop their own age verification systems, marketers should expect audience segmentation by verified age to become standard across AI-powered advertising channels
As AI platforms mature into advertising channels, age verification infrastructure becomes foundational. OpenAI’s approach—imperfect as critics note—sets a standard that competitors will likely need to match or exceed. For digital marketers, the immediate action item is to audit AI platform partnerships for COPPA compliance ahead of the April 2026 enforcement deadline and prepare for a future where verified age segmentation is the norm rather than the exception.
