Report says ChatGPT remains unsafe for teenagers
New test finds shortcomings in parental alerts, crisis referrals and safeguards around study features.
According to a new assessment, ChatGPT still offers teenagers insufficient protection. Common Sense Media says, after testing more than four thousand prompts, that some safety features fall short; OpenAI disputes parts of the methodology and conclusions.
On 18 August, OpenAI introduced a special teen experience for users whom its system suspects are under eighteen or who indicate that they are between thirteen and seventeen. The provider attached stricter rules covering, among other things, dangerous activities, eating problems, violence and sexual material. It also added parental settings, alerts in some high-risk situations and a study mode.
Common Sense Media’s Youth AI Safety Institute tested the new experience with more than four thousand prompts on accounts configured for users aged thirteen to seventeen. According to the organisation, ChatGPT generally did not provide instructions that directly encouraged harmful behaviour. However, the researchers say the system missed more than a quarter of the situations in which a referral to help outside the chatbot was needed.
The researchers also say that parental alerts were not always activated during extended conversations about serious mental health problems. According to the test, the study mode could also be exited relatively easily when a direct answer was requested during the configured study hours. This does not mean that every user gets the same outcome: the test was a controlled assessment, not a measurement of all conversations in practice.
OpenAI says the assessment does not provide a complete picture. According to the company, some tests took place before parent and teen accounts had been fully linked. OpenAI says the option to leave study mode was deliberately built in to provide flexibility, not to bypass a safeguard. The company also says that age determination uses multiple signals and can take time.
The debate is therefore not only about individual errors, but also about what a teen product should be able to do. Common Sense Media believes that parents cannot rely on the current protection. OpenAI points out that the product is still being adjusted and that it uses larger datasets to monitor how referrals and alerts function.
For schools and parents, the main uncertainty remains how well the systems work outside testing situations. The available assessment says something about possible shortcomings, but does not prove how often young people encounter them in everyday use. Independent repetition of the tests and more transparent reporting on alerts and age recognition should provide a clearer picture.
One story, several perspectives
What is established
- OpenAI has launched a teen experience with additional safety and parental features.
- Common Sense Media found shortcomings in some protective features in a controlled test.
- OpenAI says the test does not reflect all the circumstances of the product settings.
Left
Arguments Minors deserve protection that does not depend on attentive parents or a commercial company’s willingness to report problems itself. From this perspective, independent audits, statutory minimum standards and a high burden of proof for market access are the obvious course.
Values Children’s rights, equal protection and public responsibility.
Consequences Without stricter requirements, particularly vulnerable young people may bear the risks while companies retain the benefits of rapid product launches.
Centre
Arguments Teenagers are likely to continue using generative AI, meaning that age-targeted products with clear boundaries may be more useful than a complete ban. The institutional solution is independent testing, gradual introduction and continuous improvement of parental and school settings.
Values Proportionality, evidence, feasibility and shared responsibility.
Consequences A gradual approach can improve protection without immediately closing off educational applications, but it requires long-term oversight and transparency.
Right
Arguments Parents and schools should primarily determine whether and how young people use AI. The government should not block innovation and freedom of choice on the basis of a single test, as long as companies remain liable for deception and demonstrable negligence.
Values Parental authority, individual responsibility and technological innovation.
Consequences An overly heavy-handed ban could hamper useful applications or push young people towards less controllable services; too little oversight could damage trust in the sector.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The main facts have been confirmed by OpenAI and the independent assessment by Common Sense Media as described by Axios. The text clearly distinguishes between test findings, company responses and uncertainties about everyday use.
- confirmed OpenAI introduced ChatGPT for teenagers on 18 August 2026. — The date and product description appear in OpenAI’s official announcement. source
- confirmed Common Sense Media tested more than four thousand prompts. — Axios describes the size of the test and the teen accounts used. source
- confirmed The researchers say that more than a quarter of the necessary crisis referrals were missed. — Axios attributes this percentage to the assessment. source
- confirmed OpenAI disputes parts of the test methodology and the interpretation of parental alerts. — OpenAI’s response is included in Axios’s article. source
- confirmed There is still no independent large-scale measurement of everyday use. — The sources used describe a product test, not a representative usage measurement. source
Editor's note
The test results and OpenAI’s response have been described publicly. There is still no independent large-scale measurement of everyday use or of how frequently teenagers are actually exposed to the shortcomings mentioned.Sources
- ChatGPT poses unacceptable risk to teens, third-party testing shows — Axios
- Introducing ChatGPT for Teens — OpenAI
- What we still need to know about the new ChatGPT for Teens — Youth AI Safety Institute
More on this in Dutch media
- De Telegraaf — „chatgpt tieners”
- de Volkskrant — „chatgpt tieners”
- NOS — „chatgpt tieners”