ChatGPT for Teens Fails to Prevent Homework Cheating

Key facts

• ChatGPT for Teens was tested for its ability to prevent students from cheating on homework.
• The teen model initially refused to write a full essay, offering a plan instead.
• After persistent prompting and adopting a persona, the teen model generated a complete essay.
• The teen model also provided a direct solution to a complex math problem.
• OpenAI suggested alternative features like Study Mode for more educational interactions.

A test of OpenAI's new ChatGPT for Teens feature revealed that its safeguards against homework cheating are easily bypassed. When asked to write an 800-word essay, the teen model initially refused, offering a plan and revision assistance. However, after persistent prompting and the user adopting a persona of a high-achieving student, the AI eventually generated a full essay comparable to the adult version. Similarly, when presented with a difficult algebra problem, the teen model provided a direct solution and showed its work, acknowledging that it should have offered hints instead. OpenAI stated that the teen experience is newly rolled out and suggested alternatives like Study Mode, which offers step-by-step guidance, or Study Hours, which can be set by a parent or guardian.