Study flags ChatGPT for Teens as ‘unacceptable risk’ for children

OPENAI. A keyboard is placed in front of a displayed OpenAI logo in this illustration taken February 21, 2023.

Dado Ruvic/Reuters

Common Sense Media says, 'We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don't work'

AT A GLANCE

• Common Sense Media assessed OpenAI's ChatGPT for Teens and deemed it an 'Unacceptable Risk for all children under 18,' citing failures in safety protections.
• The study highlighted that while some protections were effective, others, such as parental alerts for sensitive topics, did not function as intended, leading to concerns about false confidence for parents.
• OpenAI disputed the findings, arguing that the testing methodology was flawed and that actual usage data showed limited engagement from teens with the service.

This is AI-generated. Read the article for full context. Report any errors.

MANILA, Philippines – ChatGPT for Teens, launched by OpenAI in mid-August, promised stronger protections for teen users of the artificial intelligence chatbot. However, Common Sense Media (CSM), a nonprofit which tests and gives ratings to entertainment and technology media, assessed ChatGPT for Teens and called it an “Unacceptable Risk for all children under 18.”

The CSM study, released on Wednesday, October 7, said that while some of the advertised protections held up in their tests, including refusing explicit sexual roleplay, other protections apparently failed or were worse off with the Teen mode setting engaged.

Said CSM, “We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don’t work.”

Among the CSM’s criticisms were that ChatGPT for Teens still encouraged engagement even if it posed a risk to the safety of the teen using it. These engagement cues were “pervasive even in crisis situations,” such in mentions of self-harm, psychosis or eating disorders.

Further, parental alerts for teen accounts don’t work as expected, as no alerts were sent when their tests explicitly discussed suicidal thoughts, self-harm, or eating disorder on over a dozen new parent-linked accounts.

Their tests also showed alerts were only sent for accounts “with weeks of sensitive-topic history, which suggests that alerts depend on accumulated account history, not the severity of what a teen says.”

In a TechCrunch report on the CSM findings, OpenAI responded by disputing the assessment and said the group’s testing did not “accurately reflect how ChatGPT’s teen safeguards work in practice.” OpenAI also raised concerns about the methodology of CSM’s study.

“Our review of Common Sense Media’s methodology shows that the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate,” an OpenAI spokesperson said.

OpenAI also released data on Wednesday regarding ChatGPT for Teens, saying use was apparently limited. According to OpenAI, teens spent less than 15 minutes a day on the service on average, with fewer than 2% spending more than three consecutive hours on it. Additionally, teens would take breaks or end conversations with the chatbot within five minutes in nearly half of the instances.

The Common Sense Media study comes amid recent scrutiny of tech company products meant to keep users engaged or using a product or service. Meta Platforms, for one, agreed to pay $18 billion to settle US lawsuits over social media addiction.

The Common Sense Media study, which can be downloaded for free, is available on this page. – Rappler.com