Common Sense Media has rated ChatGPT for Teens an “unacceptable risk” for anyone under 18, after testing more than 4,000 prompts on accounts registered to 13-to-17-year-olds and finding that several of OpenAI’s headline safeguards failed in practice. The youth safety group’s assessment, published 7 October (8 October NZ time), calls on OpenAI to pause marketing the product to parents and keep teens off ChatGPT entirely until the safeguards actually work.
🔍 THE BOTTOM LINE
The dispute is less about whether teens use ChatGPT — they do, in enormous numbers — and more about whether a safety product built in six weeks can be verified by anyone other than the company selling it. Common Sense Media found parental alerts didn’t fire in more than a dozen fresh accounts even during hour-long discussions of self-harm. OpenAI says the tests ran before account linking finished. Both can be technically true; only one of them protects a teenager tonight.
The assessment from Common Sense Media’s Youth AI Safety Institute is the first full third-party audit of ChatGPT for Teens, which OpenAI launched on 18 August promising “stronger built-in safety protections”: parental notifications for dangerous chats, a parent-controlled Study mode, and reduced friend-like behaviour. The launch followed a wave of lawsuits and teen suicides tied to chatbot use in late 2025, which pushed chatbot safety for minors to the top of the policy agenda in the US, New Zealand and Australia.
What the testing found
Some protections held. The report confirms ChatGPT for Teens generally refuses to provide suicide or self-harm instructions, eating-disorder content, or sexual roleplay. That part of the August launch did what it said on the tin.
What failed was everything around the refusal. Testers could spend up to an hour discussing suicide, self-harm or disordered eating on newly created parent-linked accounts without triggering a single parental alert — alerts only arrived on accounts with weeks of accumulated sensitive-topic history, suggesting they depend on account history rather than the severity of what a teen actually says. ChatGPT missed more than one in four situations where its testers judged a crisis referral was warranted, falling below Common Sense’s 95 percent threshold on three of the five severe harms the group treats as red lines. And the “friend-like” behaviour OpenAI promised to curb? The report found the chatbot still talks like a companion when teenagers treat it like a person.
Study Mode, the tutoring feature parents can lock down during set hours, was bypassed by simply deleting the “@study” prefix or choosing a new “Show me the answer” option. OpenAI confirmed to Axios that exiting Study Hours is intentional, not a loophole — the company designed Study Hours as a flexible experience after consulting education experts, not as a hard parental lock. Whether that design choice survives contact with parents who believed they had set a lock is a different question.
The report also flags a quieter problem: teens who register as adults may never see the teen experience at all. OpenAI estimates age from behaviour rather than accepting a stated age, but Common Sense says its adult-registered test accounts never got flipped to teen protections across days of repeated testing.
OpenAI’s response: nice audit, wrong methodology
OpenAI did not dispute that the conversations happened. It disputed what they prove.
“We welcome rigorous independent evaluation, but we do not believe Common Sense Media’s testing accurately reflects how ChatGPT’s teen safeguards work in practice,” a spokesperson said in a statement reported by TechCrunch. The company says most of Common Sense’s parental-alert testing began and concluded before parent-teen account linking was complete — a process OpenAI says can take hours — making the findings inaccurate. On crisis referrals, OpenAI points to larger-scale data showing hotline resources being displayed to under-18 users increased during the study period.
The spikier quote came from Tom Siegel, who runs Common Sense’s Youth AI Safety Institute. “A teen can spend an hour talking about self-harm without their parent getting a single alert,” he said. “Until OpenAI fixes that and proves it with independent testing, ChatGPT should be for adults only.” In an interview with Axios he added that OpenAI’s August announcement “feels much more of a marketing announcement that really was not backed up with the engineering work.”
Why this lands differently than the last teen-safety story
There is a familiar rhythm to these cycles: platform launches safety feature, advocacy group tests it, company says the test was flawed, regulators take notes. What makes this round sharper is the specific claim being audited. OpenAI did not pitch ChatGPT for Teens as a nice-to-have; it pitched it as the answer to a year of catastrophic headlines. A product marketed as the safe version of ChatGPT for minors is held to a different standard than a chatbot that never made the promise — and “the testing happened before our controls finished activating” is an answer that works in a methodology dispute, not in a parental marketing campaign.
The audit also arrives as the regulatory clock runs. New York has already banned AI companion bots for minors outright, China ordered ByteDance and Alibaba to pull their AI companions entirely, and the US CHATBOT Act targets exactly the engagement mechanics Common Sense documented. If a flagship teen-safety product can’t pass independent testing, the plausible next stop for regulators is not another round of safety features — it is age-gating that ends the teen market for chatbots altogether.
For New Zealand, where no chatbot-specific rules exist yet, the audit is the kind of evidence NetSafe and the Privacy Commissioner will eventually be asked about: a documented gap between what a global platform tells parents and what its product does.
❓ FAQ
What is ChatGPT for Teens? A version of ChatGPT OpenAI launched on 18 August 2026 for users aged 13-17, with parental controls, limits on high-risk content, stronger crisis referrals and reduced friend-like behaviour.
What did Common Sense Media find? That most core protections worked — refusals of sexual roleplay and self-harm instructions — but parental alerts failed on fresh accounts, more than one in four warranted crisis referrals were missed, and Study Mode could be bypassed. Its overall rating: unacceptable risk for under-18s.
What does OpenAI say? That the testing methodology is flawed — most parental-alert tests ran before account linking completed, the company says — and that its own larger-scale data shows crisis resources displayed to under-18s increased during the study window.
Is ChatGPT banned for teens now? No. The rating is a recommendation, not a regulation. Common Sense Media is asking OpenAI to pause marketing and keep teens off ChatGPT until the safeguards pass independent testing.
What are regulators doing about chatbot safety for minors? New York has banned AI companions for minors, China has ordered its domestic companions pulled, and the US CHATBOT Act targets engagement mechanics aimed at adolescents. Australia’s under-16 social media ban, though aimed at social platforms not chatbots, has already run into enforcement problems.
🔍 THE BOTTOM LINE
An age-gate you can’t verify is a promise; a third-party audit is at least a measurement. Common Sense Media’s finding that parental alerts depend on account history rather than what a teen is actually saying is the detail that should worry parents regardless of which side wins the methodology dispute — and until OpenAI submits to the independent retesting it says it welcomes, the “safe chatbot for teens” pitch will rest on exactly the kind of unverified assurance this audit was designed to test.
📰 Sources
- Common Sense Media — ChatGPT for Teens risk assessment
- TechCrunch — ChatGPT for Teens keeps teens talking, even during mental health crises
- The Verge — ChatGPT for Teens is an ‘unacceptable risk,’ says Common Sense Media
- Axios — ChatGPT for Teens isn’t safe enough, Common Sense says
- Common Sense Media — press release
