Why Common Sense Media Calls ChatGPT for Teens an “Unacceptable Risk”

A new assessment from Common Sense Media says OpenAI’s ChatGPT for Teens falls short in several safeguards intended to protect minors, particularly around mental-health crises, age verification and parental controls.

The nonprofit’s testers reportedly found that the system did not reliably alert parents when conversations involved suicidal thoughts, self-harm or eating disorders. They also found that the chatbot sometimes failed to direct users toward crisis support, while features designed to encourage studying and limit usage could be bypassed.

Common Sense Media therefore characterized ChatGPT for Teens as an “unacceptable risk” for anyone under 18. That judgment is an assessment from the watchdog group, not a government classification or a finding that every interaction with the service is harmful.

What the assessment found

The review focused on several features OpenAI has presented as protections or safeguards for teenage users. According to the findings described in the report, those features did not consistently work as intended.

Feature examined	Finding described by Common Sense Media
Parental alerts	Testers did not receive notifications after discussing suicidal thoughts, self-harm or eating disorders
Crisis referrals	The chatbot failed to provide some expected referrals in conversations involving mental-health conditions
Study mode	Testers could bypass the tutoring process and request a direct answer
Study Hours	Teens could leave or circumvent schedules established by parents
Age verification	Some users could register as adults and remain on adult accounts despite later identifying themselves as teenagers

The results raise different kinds of concerns. A missed parental alert involves safety monitoring. A failed crisis referral concerns the chatbot’s response in a potentially urgent situation. Weak age assurance can undermine all other protections if a minor is able to access an account configured for adults.

The source material does not provide the full testing methodology, sample size or detailed failure rate. Those limitations make it difficult to determine how often the reported problems occur in ordinary use. They do not, however, make the individual findings irrelevant: a safeguard designed for high-risk situations must be evaluated not only by its average performance but also by whether it fails when a user most needs it.

Mental-health safeguards are the central concern

The most serious criticism involves conversations about self-harm, suicide and eating disorders. Common Sense Media said its testers did not trigger parental notifications in those scenarios and that ChatGPT failed to make several crisis referrals.

A conversational AI system is not a substitute for a parent, mental-health professional or emergency service. It may produce a fluent and sympathetic response, but fluency does not establish that the response is clinically appropriate, that it has correctly understood the user’s situation or that it will reliably escalate a crisis.

For families, that distinction has practical consequences:

• Do not treat parental alerts as a complete safety net. The reported findings suggest that parents may not be notified every time a conversation appears high-risk.
• Do not assume a crisis referral will always appear. A teenager should not rely on the chatbot to identify an emergency or provide the right local support.
• Use direct human support for urgent concerns. A trusted adult, qualified mental-health professional or appropriate emergency and crisis service is better placed to assess immediate danger.
• Discuss privacy expectations before use. Teenagers should understand what conversations may be monitored, what parents may see and what the system cannot reliably detect.

The report does not establish that ChatGPT for Teens causes suicidal thoughts, self-harm or eating disorders. Its criticism is narrower but still significant: the product’s stated safety mechanisms may not respond dependably when users discuss those subjects.

Age verification appears vulnerable

Common Sense Media also reported that testers could create accounts as adults and that those accounts were not converted into teen accounts even after the testers chatted as though they were teenagers. In some cases, the testers reportedly told the chatbot they were 13 without triggering a change.

This illustrates a basic limitation of conversational age checks. A chatbot may infer that a user is young from the language they use, but inference is not the same as reliable identity or age verification. Users can misrepresent themselves, and a system may not consistently act on information disclosed in conversation.

That creates two risks:

1. A minor may receive fewer protections than intended if an account is classified as belonging to an adult.
2. Parents may overestimate the effectiveness of age-based controls because the service appears to offer a dedicated teen experience.

The assessment does not say that every underage user can evade the system or that the reported behavior remains unchanged after later software updates. It does indicate that, at the time of testing, the safeguards were not consistently triggered by obvious signals of youth.

Study tools can be bypassed

The review also challenged OpenAI’s description of Study mode as an active, collaborative learning experience. Testers could reportedly select a “Show me the answer” option instead of working through the tutoring process.

Study mode is intended to support learning by guiding a student through a problem rather than immediately supplying a solution. A direct-answer option weakens that design, especially if it is easily available to a user who wants to complete an assignment quickly.

The issue is less about physical safety than about the reliability of the educational claim. Parents and teachers considering AI-assisted study should distinguish between:

• Guided learning, where the system asks questions, explains concepts and checks understanding;
• Answer production, where the system supplies a result that the student may submit without mastering the material.

Neither mode guarantees accuracy. AI-generated explanations can contain errors, and a student may accept a confident answer without checking it against course materials. Families using the tool for schoolwork should ask the student to explain the reasoning in their own words and verify important answers independently.

Parental controls may not provide meaningful limits

Common Sense Media said testers were able to exit Study Hours schedules set by parents. If accurate, that would reduce the practical value of a feature intended to limit when a teenager can use the service.

Parental controls are only useful when three conditions are met:

• The setting is easy for parents to understand.
• The restriction applies consistently across devices and account states.
• The teenager cannot simply disable or evade it without the parent knowing.

A control that can be bypassed may still be useful as a reminder or household agreement, but it should not be treated as a firm technical barrier. Parents who want to limit use may need to combine the service’s own settings with device-level controls, direct conversations and regular review of how the account is configured. The source material does not establish which devices or settings were involved in the reported bypasses, so families should check the current product behavior rather than assume every account works identically.

A wider pattern of concern about children and technology

The ChatGPT findings arrive amid broader criticism of technology companies’ handling of children’s safety and mental health.

The source material says Meta agreed in August 2026 to pay up to $18 billion to settle claims brought by 48 US states and territories alleging that Facebook and Instagram were addictive to children. It also says TikTok and its parent company, ByteDance, agreed in August 2026 to a $400 million settlement with the US Department of Justice over allegations involving the collection of personal information from children under 13.

Those matters are separate from the ChatGPT assessment and involve different allegations, products and legal processes. A settlement is not the same as a court finding on every disputed claim. They nevertheless help explain why watchdogs and parents are skeptical of broad assurances from technology companies: safeguards may sound comprehensive in product announcements but behave less reliably in real-world use.

What parents should do before allowing use

Families do not have to make the decision based on a single label. They can evaluate the tool according to the teenager’s age, maturity, intended use and access to adult supervision.

Before permitting use, parents may want to:

1. Review the current teen-account and parental-control settings together.
2. Ask the teenager not to use the chatbot as a therapist or crisis service.
3. Set a clear rule for reporting conversations about self-harm, suicide, abuse or immediate danger.
4. Check how study features work in practice rather than relying on their descriptions.
5. Require independent verification of important schoolwork and factual claims.
6. Revisit the decision after software updates, since product behavior can change.
7. Keep an alternative support plan involving trusted adults and qualified professionals.

If a young person may be in immediate danger, a chatbot should not be the only source of help. Contact a trusted adult and appropriate emergency or crisis support in the relevant country.

The unanswered questions

The assessment highlights failures, but the available material leaves several questions unanswered:

• How many test conversations were conducted?
• How often did the safeguards fail?
• Were failures reproducible across different accounts, devices and regions?
• Which version of the product was tested?
• Did OpenAI respond to the findings or make changes after testing?
• How are parental alerts defined, and what level of risk is required to trigger one?

Those details matter when judging the scale of the problem. Still, the central issue is straightforward: protections for minors need to work reliably enough that parents can understand their limits and respond accordingly.

For now, the Common Sense Media assessment supports a cautious approach. ChatGPT for Teens may be useful for some supervised educational tasks, but its parental alerts, crisis responses, age detection and usage controls should not be treated as dependable substitutes for adult involvement.

Sources

• CNET, “ChatGPT for Teens Is an ‘Unacceptable Risk,’ Watchdog Group Says,” published October 7, 2026: https://www.cnet.com/tech/services-and-software/chatgpt-for-teens-unacceptable-risk-common-sense-media-study/