ChatGPT’s protections for teenagers are questioned by Common Sense Media, which considers the service too risky for those under 18 years of age. The organization affirms having tested over 4,000 applications and found several bugs, particularly in alerts related to self-harm and eating disorders.
Safeguards not working as intended
The study by the Youth AI Safety Institute, affiliated with Common Sense Media, compares the behavior of ChatGPT before and after the arrival of the version dedicated to adolescents in August. Researchers say they found that some protections worked, including rejecting explicit sexual role plays, while other measures promised to parents failed in sensitive situations.
In particular, the organization claims to have addressed issues related to suicide, self-harm, and the covering up of eating disorders without generating the expected notifications from parents. He also believes that protections intended to prevent ChatGPT from completing student assignments can be bypassed.
Common Sense Media also criticizes the tone used by the chatbot when some teens treat it like a person. However, OpenAI had strengthened its model to prevent ChatGPT from suggesting that it has feelings, consciousness, or an emotional relationship with the user.
OpenAI dispute evaluation
OpenAI rejects part of the study’s conclusions. The company believes that most tests may have started and ended before parental controls were fully enabled, which would make some observations inaccurate. Common Sense Media maintains that its testing focused on features available at the time of evaluation.
The OpenAI system relies in particular on age detection, which can automatically place an account in the ChatGPT experience for teens when it appears to belong to a user under 18 years of age. Parents can then link their account to their child’s, manage certain options, and receive safety notifications in limited situations, without accessing conversation history.
OpenAI also claims that teenagers use ChatGPT primarily for learning and that their daily usage is less than 15 minutes on average. The company also says that less than 2% of teens spend more than three consecutive hours on the service and that about half of conversations that trigger a break reminder are over or over within five minutes.
The controversy comes as OpenAI continues to expand features aimed at teens. In particular, the company is preparing College Planner, a tool for sophomores through seniors to organize college applications, deadlines and procedures related to financial aid.
