AI firms add teen safeguards as OpenAI, Meta, Anthropic clash on access

On Thursday, OpenAI published an official stance on teen access to AI, arguing that young people will be less prepared for adulthood if they don't practice with the technology while they are young. Lauren Jonas, OpenAI's head of youth well-being, framed the approach as learning from past industry mistakes: "The principle here is to avoid the mistakes that were made before us." She distinguished AI from social media, noting that teens primarily use AI tools for schoolwork. OpenAI's proposed safeguards include nudges to take breaks and time limits that parents can set.

The company's position contrasts sharply with Anthropic's stricter gate: Anthropic requires users to enter a birthday confirming they are older than 18 before accessing its AI products. Meta has taken a middle path, announcing it will notify parents if their child discusses self-harm with its chatbot—moving closer to OpenAI's supervised-access model while adding a direct safety trigger.

All three companies face a practical barrier: teens have little incentive to submit their real ages or connect a parent's account voluntarily. To address this, major AI labs have begun using machine learning to predict user ages based on query patterns, automatically flagging accounts for additional verification. The article identifies this age-prediction method as the most sophisticated protection approach available so far. However, the piece closes with a sobering historical reminder: the past decade has shown that teen safety depends on more than product updates alone. Sustained protection requires support from communities, schools, parents, and the teens themselves—a lesson social media platforms have struggled to implement at scale.

Frontier AI labs face mounting pressure to demonstrate teen safety, but the industry is splitting on philosophy. OpenAI's position—that teens should use AI under parental supervision to build skills for adulthood—reflects a belief that abstinence from new technology is itself a risk. Anthropic's age gate is simpler but blunter. Meta's self-harm notification sits between them: it allows teen access but triggers a safety signal to parents.

The core tension is enforcement: teens have little motivation to submit real ages or connect parent accounts if they can access AI tools without doing so. To address this, companies are turning to machine learning to infer age from user behavior and prompt verification. Yet the article hints at a deeper lesson from the social media era—that product features and age gates alone have never been sufficient. Teen safety historically required alignment across parents, schools, and communities, not just platform rules. The question is whether AI companies can make that ecosystem work where social media platforms largely failed.