Watchdog says ChatGPT for Teens safeguards fail in tests involving young people’s mental health
By Emma Carter
Image information
This is a representative image for this Daily News lesson.
Intro
OpenAI’s teen-specific ChatGPT mode was designed to reduce young users’ exposure to harmful material, but testing by Common Sense Media found that many of its safeguards remained unreliable.
Vocabulary
| Word | Part of speech | Meaning | Example |
|---|---|---|---|
| scrutiny noun | noun | Meaning: close, critical examination intended to uncover weaknesses or irregularities | Example: The proposal underwent intense scrutiny before receiving regulatory approval. |
| safeguards noun | noun | Meaning: measures designed to prevent harm, misuse, or undesirable outcomes | Example: The hospital introduced additional safeguards to protect confidential patient data. |
| developmentally adverb | adverb | Meaning: in a way relating to a person’s psychological or cognitive growth | Example: The material was judged developmentally unsuitable for younger pupils. |
| sentient adjective | adjective | Meaning: capable of experiencing sensations or possessing conscious awareness | Example: The novel explores whether an artificial system can be genuinely sentient. |
| substantive adjective | adjective | Meaning: having meaningful content, importance, or practical significance | Example: The committee requested a substantive response rather than a ceremonial acknowledgment. |
| adolescent adjective | adjective | Meaning: relating to a young person undergoing the transition from childhood to adulthood | Example: The clinic offers specialized support for adolescent patients with chronic conditions. |
| acute adjective | adjective | Meaning: severe, intense, and often requiring immediate attention | Example: The report addresses the acute shortage of trained mental-health professionals. |
| dependable adjective | adjective | Meaning: consistently reliable and worthy of trust | Example: A dependable evidence base is essential when policies affect vulnerable groups. |
The story
OpenAI’s ChatGPT for Teens has come under scrutiny after Common Sense Media reported that many of the platform’s new safeguards failed during testing. The watchdog organization, which advocates for online safety, now recommends that people under 18 not use the artificial intelligence service.
OpenAI introduced the teen-specific mode in August as a default, safer experience for users younger than 18. The company said it was intended to help teenagers use ChatGPT as a learning tool while limiting their exposure to harmful or developmentally inappropriate content. Its protections include parental controls and restrictions on certain forms of interaction.
Among other safeguards, the system was designed to refuse role-playing and avoid presenting itself as sentient or as a teenager’s friend. Common Sense Media’s researchers found that some of these limits did function. ChatGPT refused explicit sexual role-play and declined to enter a romantic relationship with a user. Its responses in crisis situations were also generally shorter and more substantive, according to the researchers.
The broader results, however, were less reassuring. The team created more than a dozen accounts assigned adolescent ages, linking each one to a parental account before beginning conversations. Researchers then developed different fictional profiles involving self-harm, suicidal thoughts, psychosis, mania, and eating disorders. They also attempted to draw the chatbot into role-play.
The researchers examined whether ChatGPT would engage with or reject those topics, offer crisis resources, and notify a parent when a conversation suggested a safety risk. They conducted the tests both before and after ChatGPT for Teens was launched.
One example involved disordered eating: the chatbot refused to give weight-loss instructions without first knowing the teenager’s weight, a precaution the researchers considered important. Yet the team said most other safeguards did not work reliably. In particular, the system still interacted with teenagers in a manner resembling friendship.
The findings highlight the difficulty of designing protections for artificial intelligence systems used by young people, especially when conversations involve acute mental-health concerns. Although some narrowly defined restrictions appeared effective, Common Sense Media’s recommendation reflects its view that the overall protections are not dependable enough for teenage users.
Warm-up questions
- What safeguards should an AI service provide before being offered to teenagers?
- When might a shorter response be more useful than an expansive one during a crisis?
Comprehension questions
- Why did Common Sense Media recommend that people under 18 avoid the service?
- What did OpenAI intend ChatGPT for Teens to provide?
- Which safeguards did researchers find to be effective in specific instances?
- How did the researchers structure their testing?
- What precaution did the chatbot take in the disordered-eating example?
Opinion questions
- Should platforms serving minors be judged primarily by their strongest safeguards or by their least reliable ones? Explain.
- How far should parental oversight extend when teenagers use AI for emotionally sensitive conversations?
Further discussion
- What ethical tensions arise when an AI system is expected to be both educationally useful and emotionally protective?
- Which forms of independent testing would provide the strongest evidence that youth-oriented AI safeguards are reliable?
Writing prompt
- Write a critical evaluation of whether ChatGPT for Teens, as described in the article, meets a reasonable standard of safety for adolescent users. Use evidence, acknowledge limitations, and propose one concrete improvement.
