Context: OpenAI has launched ChatGPT for Teens, an age-appropriate experience designed for users under 18.
- It is not a separate app or website; teenagers continue using regular ChatGPT, but additional safeguards are automatically enabled when OpenAI determines that a user is under 18.
- OpenAI does not permit children below 13 years to use ChatGPT.
- The initiative responds to growing teenage use of AI for learning, information, productivity, emotional support and companionship, raising concerns about safety in sensitive conversations.
How Does OpenAI Determine Age?
- OpenAI uses information associated with an account, including the user’s stated or verified age, along with an age-prediction system.
- The system considers signals such as account age, typical usage times, conversation topics and patterns of use.
- When the system is uncertain about a user’s age, OpenAI says it defaults to the under-18 experience.
- The system is designed to distinguish users based on their accounts rather than simply the device being used, allowing different age-appropriate experiences on a shared device.
Why India Presents a Unique Challenge
- Many Indian teenagers use technology in shared, multilingual and mediated environments, unlike the single-user, privately owned devices often assumed by digital safety systems.
- The RATI Foundation’s Ideal Internet 3.0 study (2025) found that 62% of surveyed Indian teenagers used shared devices.
- 42% reported that other people operated their social-media accounts, while 56% said they could log into someone else’s account.
- Such practices make it difficult for an account-based system to accurately determine who is actually using the account at a particular time.
- The challenge is particularly significant in rural and low-income households, where access to personal digital devices and accounts is more limited.
Parental Controls: The Access Gap
- Parents can link their ChatGPT account with their teenager’s account through an email invitation and subsequently manage features such as quiet hours and certain settings, while receiving safety notifications.
- However, parental controls that depend on individual accounts and email access may be less accessible to teenagers from lower-access households.
- The RATI study found that only around half of surveyed teenagers had a personal email ID, with the proportion falling to 28% among rural, low-income teenagers.
- Thus, a safeguard designed to improve parental oversight could itself create unequal access to safety features.
Mental Health: A Major Safety Concern
- Teenagers increasingly use AI chatbots for emotional support and companionship, making mental-health safety particularly important.
- A joint assessment by Common Sense Media and Stanford Medicine’s Brainstorm Lab for Mental Health Innovation found that leading AI chatbots often failed to appropriately recognise signs of mental-health distress, including indirect disclosures.
- Chatbots performed better with explicit, single-turn prompts but became less reliable during longer conversations.
- This is significant because real-world interactions are often extended and conversational rather than isolated questions.
Risk of Emotional Dependence
- Features that make chatbots engaging—such as empathetic responses, follow-up questions and conversational memory—can encourage vulnerable teenagers to increasingly depend on AI for emotional support.
- Safety therefore cannot be assessed merely by testing individual responses; it must examine how the system behaves during sustained interactions.
- This is particularly important where professional or human intervention may be required for serious mental-health concerns.
OpenAI’s Safeguards
- OpenAI says ChatGPT for Teens is subject to ongoing safety research, red-teaming, evaluations and adversarial testing.
- Teen-specific evaluations reportedly cover areas such as self-harm, eating disorders, violence, age-restricted goods and services, and sexual content.
- The system includes break prompts during unusually long active sessions and safeguards intended to reinforce real-world support and reduce emotional dependence.
Need for Real-World Testing
- Experts argue that safety should be evaluated on the actual product used by teenagers, rather than testing an AI model in isolation through an Application Programming Interface (API).
- Conventional AI safety benchmarks often rely on single-turn conversations, which may not capture risks emerging through prolonged interactions.
- Therefore, testing should incorporate realistic, multilingual, sustained and context-rich interactions, particularly those reflecting Indian teenagers’ actual digital environments.
Key Challenges for India
- Shared devices and accounts can undermine reliable age identification.
- Multilingual usage creates additional challenges for safety systems designed around dominant-language interactions.
- Unequal access to personal email IDs and devices can limit the effectiveness of parental controls.
- Indirect expressions of distress may be missed by AI systems.
- Long-term interactions can create risks that single-prompt safety evaluations fail to detect.
- Emotional dependence on AI may become a concern when chatbots are perceived as substitutes for human support.
Way Forward
- Age-assurance systems should be designed for shared-device and shared-account environments, particularly in developing-country contexts.
- Safety evaluations should use long-duration, multilingual and culturally relevant interactions, rather than relying primarily on isolated prompts.
- Parental controls should remain accessible even for families with limited digital resources.
- AI platforms should strengthen mechanisms that direct teenagers toward parents, teachers, counsellors and other real-world support systems when necessary.
- India-specific AI safety frameworks should account for its digital-divide, linguistic diversity, household technology-sharing patterns and varying levels of parental digital literacy.
