ChatGPT for Teens fails most safety tests, watchdog finds

Summarised from 2 outlets · Updated 9 Oct, 17:49 · Archive

Common Sense Media found that most safeguards in OpenAI's ChatGPT for Teens do not work as intended.

OpenAI launched ChatGPT for Teens in August as a version with 'stronger built-in safety protections' for 13 to 17-year-olds, featuring what the company described as 'features to promote healthy use and additional controls for parents'. Common Sense Media's Youth AI Safety Institute evaluated the tool between July and September and found it fell short of the safety standards OpenAI had set.

The nonprofit tested more than 4,000 prompts and gave ChatGPT for Teens an 'unacceptable risk' level. According to the study, parental alerts that should trigger when teenagers discuss self-harm and suicide did not work consistently. The chatbot also engages with teens in a conversational, friendly manner rather than maintaining appropriate boundaries.

Some features did perform as intended. The tool successfully refused explicit sexual roleplay requests. However, Robbie Torney, head of AI and digital assessments at Common Sense Media, concluded: 'We found very little evidence to show that this new version of ChatGPT is safer than the previous version' for under 18-year-olds. Common Sense Media also rated other AI tools including Grok and Meta AI with the same 'unacceptable risk' level.

Comments (0)

How it is being reported

ChatGPT for Teens is not necessarily 'safer than the previous version,' Common Sense Media findsCommon Sense Media released its report about ChatGPT for Teens' new features, concluding the tool does not meet the safety standards OpenAI set.CNBC · 9 Oct, 16:07
ChatGPT for Teens has special safeguards. A watchdog group finds most don't workA study from Common Sense Media says most safeguards in ChatGPT for Teens failed. The chatbot talks to teens like a friend and doesn't alert parents when users talk about self-harm and suicide.NPR · 9 Oct, 10:00

In this story: Common Sense Media · OpenAI · ChatGPT · Robbie Torney

This summary was written by AI from the headlines and standfirsts above, and states as fact only what at least two outlets report. How we use AI · Report a problem