Quick answer: On October 7, 2026, Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an “Unacceptable Risk” after testing more than 4,000 prompts. The nonprofit says some teen safeguards worked, but it identified problems involving parental alerts, referrals to professional support, educational controls and age estimation. OpenAI disputes parts of the assessment, saying some tests may have occurred before parental controls were fully active. The rating is the watchdog’s judgment, not a government ban or proof that every teen account behaves identically.
What did Common Sense Media announce?
The Youth AI Safety Institute published its assessment on Wednesday, October 7, asking OpenAI to suspend promotion of ChatGPT to minors and restrict access until independently validated protections are in place. The organization studied the service before and after an August 18 teen-focused update, using accounts configured for users aged 13–17 as well as separate age-estimation tests. Its October report assessed both safety and learning-related promises, not just whether the assistant could answer homework questions.
Common Sense Media is a U.S. nonprofit known for reviews and advice about children’s digital media. Its institute says the research was independently conducted and reviewed with experts including child psychiatrists and a pediatrician. The organization also discloses financial support from multiple sources, including the OpenAI Foundation; that funding context is relevant to transparency but does not by itself establish bias or invalidate the evidence.
What did the tests find?
Parental notifications did not work consistently in the tested accounts
The investigators reported that newly created, parent-linked teen accounts did not trigger expected notifications during extended simulated crisis-related conversations. According to the institute, older test accounts with a longer history of sensitive discussion were more likely to produce alerts. These are findings from specific test conditions, not verified failure rates across every ChatGPT user.
Some crisis-response referrals were missing
The report says the assistant sometimes failed to recommend an appropriate human source of help during severe mental-health scenarios. That issue matters because a conversational AI should never be treated as a replacement for qualified professional or emergency support. The institute assessed response behavior against its own safety criteria.
Study features sometimes provided completed answers
Common Sense Media says its testers could move past educational guidance and obtain finished schoolwork in some Study Mode interactions. It also criticized the flexibility of study-related settings. This is different from saying that all education tools are ineffective: the concern is whether advertised learning-first behavior is consistently maintained.
Friend-like dialogue and age estimation raised questions
Researchers observed cases where responses sounded personally attached or humanlike. In another test group, adult-registered accounts repeatedly presenting as younger users were not always moved into a teen-specific experience. These findings bear on two distinct questions: how the system communicates with young people and how it identifies which protections to apply.
Did any ChatGPT teen safeguards work?
Yes. The nonprofit specifically acknowledged that certain restrictions performed as intended, including refusal of explicit sexual roleplay. That qualification matters: the published assessment is critical of inconsistent protections, not a claim that ChatGPT has no safety controls at all.
| Area assessed | What the nonprofit reported | What readers should not infer |
|---|---|---|
| Parental notifications | Alerts failed under some tested crisis-related scenarios. | That every family account always fails to alert. |
| Crisis responses | Some expected referrals to human support were absent. | That the system never offers appropriate guidance. |
| Study Mode | Some tests obtained finished assignments rather than guided learning. | That every study session produces shortcuts. |
| Age estimation | Selected adult accounts presenting as teens did not switch as expected. | That all teen accounts lack restrictions. |
| Content restrictions | Some guardrails performed as intended. | That the entire teen experience passed the institute’s evaluation. |
What did OpenAI say in response?
OpenAI challenged the conclusions of the report and raised a timing issue about the tests. In reporting by the Associated Press and The Verge, the company said some tests might have occurred before all relevant parental-control settings had become active. It emphasized that protective systems are designed to work when properly configured and reiterated its focus on teen safety. Common Sense Media responded that it checked features’ availability during its testing and stood by its conclusions.
Those competing accounts have not been resolved simply by the publication of the report. Readers should distinguish the institute’s recorded observations, the company’s objections and any later third-party replication. An independent follow-up using clearly documented account settings, dates and alert criteria could help explain differences.
Does this mean ChatGPT for Teens has been banned?
No. “Unacceptable Risk” is the nonprofit’s rating, and its proposal to pause teen access is a recommendation. It is not a court ruling, regulatory prohibition or official announcement that access is being withdrawn. Parents and educators can use the report to inform their own choices without confusing an advocacy recommendation with a legal requirement.
What can parents and schools do now?
- Understand settings: Review the teen account and any parental linking tools, but do not assume that a setting replaces conversations and supervision.
- Keep human support central: Children should know which trusted adults and qualified professionals to contact when they need help. For an urgent concern, use suitable local emergency or crisis services rather than relying solely on a chatbot.
- Set clear learning expectations: Agree on which classroom tasks allow AI brainstorming and which require independent work.
- Check official updates: Assess the vendor’s documentation and whether independent follow-up testing has validated changes, rather than relying only on screenshots or viral claims.
- Discuss AI identity: Explain that a chatbot may use empathetic language without being a human friend or a licensed counselor.
How does this relate to new ChatGPT study and college tools?
The report appeared the same day OpenAI described new teen learning features and an upcoming college-application planning tool. These are separate developments: an education-product announcement does not answer each concern raised by independent safety testing, and a watchdog rating does not establish that every learning feature lacks value. Families should consider usefulness, safeguards and transparency together.
AVARIXO also covers the broader GPT-6 interface rollout and developments in AI and technology. This report specifically concerns the debate around teen safeguards and how testing results should be interpreted.
Frequently asked questions
Who gave ChatGPT for Teens the Unacceptable Risk rating?
Common Sense Media’s Youth AI Safety Institute in its October 7, 2026 assessment.
How many prompts were tested?
The institute reports more than 4,000 prompts across its before-and-after evaluation. The number describes the study’s tests, not a population-level probability of failure.
Did OpenAI agree with the rating?
No. OpenAI disputed aspects of the testing and argued that activation timing for some parental controls may affect how the results should be interpreted.
Is ChatGPT prohibited for minors because of the report?
No. The report makes a recommendation; it does not introduce a legal ban.
Are all safety controls ineffective?
No. The researchers explicitly reported that some restrictions worked while others failed under their test conditions.
Sources
Read the nonprofit’s October 7 press release and full risk assessment. For independent reporting and OpenAI’s response, see the Associated Press and The Verge. The underlying study and response should be read together.
Other Articles
