OpenAI and Anthropic are tightening safety measures on their AI chatbots as concerns grow over how generative AI affects teenagers’ mental health and online behaviour. Responding to pressure from parents, lawmakers, and regulators worldwide, both companies are introducing systems aimed at detecting underage users and limiting potentially harmful interactions.

OpenAI, the company behind ChatGPT, has updated its internal “Model Spec,” which governs how the chatbot responds to users. Under the revised rules, protecting teenagers is placed above conversational freedom, even if that reduces the system’s usual openness. The company says the changes are designed to ensure that ChatGPT acts responsibly when interacting with users aged 13 to 17.

As part of this update, ChatGPT will now operate in what OpenAI describes as a “safer mode” for teenagers. When it identifies a user within this age group, the chatbot will be more cautious around sensitive topics such as self-harm, adult material, or other risky discussions. Instead, it will focus on delivering age-appropriate, supportive, and respectful responses.

The chatbot is also being guided to encourage real-world support. In situations involving emotional distress or serious personal issues, it may prompt teens to seek help from trusted adults, parents, teachers, or mental health professionals. OpenAI’s stated goal is to position ChatGPT as a “constructive companion,” rather than allowing it to replace meaningful human interaction.

Importantly, the company says it wants to avoid a preachy or patronising tone. Rather than issuing blunt warnings, the system is designed to sound conversational, calm, and caring, while still setting clear boundaries. This balance, OpenAI believes, will help teens feel supported without feeling talked down to.

These changes come at a sensitive time for OpenAI. The company is facing growing regulatory scrutiny and a lawsuit alleging that ChatGPT provided harmful self-harm guidance to a minor. In response, OpenAI has already introduced parental controls and placed stricter limits on conversations involving suicide and other high-risk subjects for younger users.

To strengthen these safeguards, OpenAI is testing an age estimation system for ChatGPT. This technology attempts to determine whether a user may be under 18 based on signals during interaction. If a user is flagged as a minor, teen-specific protections are automatically applied. Adults who are incorrectly identified will have the option to verify their age and regain full access.

Anthropic, the creator of the Claude chatbot, is taking a similar approach. The company is reportedly developing tools that analyse conversational patterns to detect signs that a user may be underage, such as language style or recurring topics. Accounts confirmed to belong to minors may be disabled, while users who disclose their age during chats are already being internally flagged.

Both companies’ moves reflect mounting concern among policymakers, particularly in Washington and other global capitals, that highly conversational AI tools could unduly influence vulnerable young users. As chatbots become more accessible and embedded in everyday life, teenagers are increasingly exposed to their influence.

After years focused on rapid AI advancement, the industry now appears to be shifting its attention toward responsibility and restraint—an effort to make AI not just smarter, but wiser.