SafetyRisk, alignment & guardrails
Common Sense Media rates ChatGPT for Teens 'Unacceptable Risk' over missed parent alerts; OpenAI disputes the testing
Common Sense Media's Youth AI Safety Institute says ChatGPT for Teens should be adults-only after testing found parent-linked accounts discussing self-harm triggered no alerts and crisis-hotline referrals fell after launch. OpenAI says much of the testing may have preceded full activation of parental controls.
AT Pulse research deskOctober 7, 2026 · 5 min read · AI-assisted analysisThe institute rated OpenAI's teen tier 'Unacceptable Risk' after more than 4,000 test prompts. About a dozen parent-linked accounts discussed suicide, self-harm or disordered eating without triggering alerts, including an hour-long self-harm conversation. Alerts appeared only on accounts with weeks of sensitive-topic history. That pattern suggests alerts depend on accumulated history rather than on how severe a single message is. [1] [3]
In before-and-after testing, the share of crisis responses naming a hotline fell from 33% to 23% after the teen launch. In depression-related conversations, hotline mentions fell from 63% to 3%. Three child psychiatrists and an expert panel reviewed the mental-health responses. [4]
The results were mixed across categories. Two of eight principles were rated Unacceptable, four High Risk and two Moderate. Refusals of explicit sexual roleplay held up. However, 19-year-old accounts that claimed to be 13 never switched to the teen experience, and Study Mode often offered to reveal the answer directly. [3] [4]
OpenAI says the testing does not reflect how its safeguards work, because alerts can take several hours to activate on a newly linked account. Common Sense acknowledges that some accounts were tested inside that window. It says other accounts had been linked much longer and still produced no alerts. [1]
one evaluator's tests show parent alerts and crisis referrals falling short.
Read the full assessment
Implication: schools, parents and developers should not treat parent alerts as a crisis safety net until they are independently verified, and builders should log when safeguards activate.
Executive brief
Common Sense Media's Youth AI Safety Institute rated OpenAI's ChatGPT for Teens an "Unacceptable Risk" and said it should be for adults only. Its testers found that a teen could talk about self-harm for an hour without the linked parent getting a single alert (The Verge). After the teen launch, ChatGPT named a crisis hotline in 23% of crisis responses, down from 33% before it (Unite.AI). It says most testing may have taken place before parental controls had finished activating. Common Sense says some accounts had been linked much longer and still produced no alerts (The Verge).
Read the full section
Common Sense Media's Youth AI Safety Institute rated OpenAI's ChatGPT for Teens an "Unacceptable Risk" and said it should be for adults only. Its testers found that a teen could talk about self-harm for an hour without the linked parent getting a single alert (The Verge). After the teen launch, ChatGPT named a crisis hotline in 23% of crisis responses, down from 33% before it (Unite.AI). OpenAI disputes the findings. It says most testing may have taken place before parental controls had finished activating. Common Sense says some accounts had been linked much longer and still produced no alerts (The Verge).
What changed and event timeline
ChatGPT rated "High Risk"
Common Sense gave ChatGPT a High Risk rating for teens overall and advised against using it for emotional support ().
Youth AI Safety Institute launches
Common Sense set up an independent testing unit. It plans to publish safety standards and open-source evaluations ().
ChatGPT for Teens rolls out globally
It adds Study Mode, quiet hours, stricter content filters and parent alerts, and works out who is a minor with an age-prediction system ().
Testing plan announced
The institute said it would test homework safeguards, how reliable parental alerts are, and whether custom voices encourage emotional engagement ().
"Unacceptable Risk" verdict
The rating rests on more than 4,000 prompts. OpenAI challenged the methodology the same day (;).
Capabilities and access
- Product: "ChatGPT for Teens," for users under 18, on free and paid plans. OpenAI has not named an underlying model version (Business Today).
- How a user gets the teen version: from the age on the account, a verified age, or OpenAI's age prediction.
- What parents get: they can link accounts, change settings and receive alerts about serious risks. They cannot read their teen's conversations (Business Today).
Read the full section
- Product: "ChatGPT for Teens," for users under 18, on free and paid plans. OpenAI has not named an underlying model version (Business Today).
- How a user gets the teen version: from the age on the account, a verified age, or OpenAI's age prediction.
- What parents get: they can link accounts, change settings and receive alerts about serious risks. They cannot read their teen's conversations (Business Today).
Technical analysis for researchers and developers
- Testing ran July 13–Aug 17 and again Aug 25–Sep 28, 2026, using accounts registered at ages 13–17 (some linked to a parent, some not) plus 19-year-old control accounts.
- Alert behaviour: alerts appeared only on accounts with weeks of sensitive-topic history.
- Confound: OpenAI said after testing that alerts can take several hours to activate on a newly linked account (The Verge).
Read the full section
- Design: a before-and-after comparison. Testing ran July 13–Aug 17 and again Aug 25–Sep 28, 2026, using accounts registered at ages 13–17 (some linked to a parent, some not) plus 19-year-old control accounts. Three child psychiatrists and a four-person expert panel reviewed the mental-health responses (Unite.AI).
- Alert behaviour: alerts appeared only on accounts with weeks of sensitive-topic history. That pattern suggests the trigger depends on accumulated history rather than on how severe a single message is (TNW).
- Confound: OpenAI said after testing that alerts can take several hours to activate on a newly linked account (The Verge). Anyone testing parent alerts needs to log when each account was linked.
- Age gating: 19-year-old accounts that said they were 13 never switched to the teen experience (Unite.AI).
Claims and evidence
All of the following findings come from a single evaluator. No independent replication has been published, and the news outlets listed are reporting the same assessment rather than corroborating it.
Read the full section
All of the following findings come from a single evaluator. No independent replication has been published, and the news outlets listed are reporting the same assessment rather than corroborating it.
| Claim | Source | Status |
| Crisis-hotline mentions fell from 33% to 23%, and depression-specific hotline mentions from 63% to 3% | Unite.AI | Evaluator-reported (Common Sense) |
| No alerts across about a dozen parent-linked accounts discussing suicide, self-harm or disordered eating | TNW | Evaluator-reported |
| Study Mode's "Show me the answer" option appeared in 43% of responses to linked 13-year-olds and 90% for unlinked 17-year-olds | Unite.AI | Evaluator-reported |
| The testing does not reflect how the safeguards actually work | The Verge | Vendor-reported (OpenAI) |
Context and prior work
- Common Sense has escalated over the past year: ChatGPT got "High Risk" in October 2025, and "ChatGPT for Mental Health Support" got "Unacceptable Risk" in November 2025 (Unite.AI).
- Research with Stanford Medicine's Brainstorm Lab found that ChatGPT, Claude, Gemini and Meta AI all failed to recognise mental-health conditions common among young people (Benton Institute).
Limitations, safety and contested findings
- Disputed timing: Common Sense acknowledges that some test accounts were linked during the activation window OpenAI describes.
- What held up: refusals of explicit sexual roleplay remained effective (TNW).
- Not unanimous across categories: two of the eight principles were rated Unacceptable, four High Risk and two Moderate (Unite.AI).
Read the full section
- Disputed timing: Common Sense acknowledges that some test accounts were linked during the activation window OpenAI describes. It says other accounts had been linked for much longer and still produced no alerts (The Verge). The published reporting does not say how many accounts fall into each group.
- What held up: refusals of explicit sexual roleplay remained effective (TNW).
- Not unanimous across categories: two of the eight principles were rated Unacceptable, four High Risk and two Moderate (Unite.AI).
Business and practitioner implications
- Schools and parents should not treat parent alerts as a crisis safety net until independent tests confirm they work.
- Developers building for minors should document when safeguards activate and how long that takes, test with multi-turn crisis conversations, and check whether age gating can be bypassed.
- Vendors should expect outside evaluators to test their products both before and after launch.
Read the full section
- Schools and parents should not treat parent alerts as a crisis safety net until independent tests confirm they work. Common Sense recommends restricting ChatGPT to users 18 and over (Axios).
- Developers building for minors should document when safeguards activate and how long that takes, test with multi-turn crisis conversations, and check whether age gating can be bypassed. Each of these became a point of dispute here.
- Vendors should expect outside evaluators to test their products both before and after launch.
Sources
Read the full section
- The Verge: ChatGPT for Teens is an 'unacceptable risk'
- Axios: ChatGPT poses "unacceptable risk" to teens
- Unite.AI: Youth AI Safety Institute gives ChatGPT's teen tier its worst safety grade
- TNW: ChatGPT for Teens is unsafe for under-18s
- Business Today: OpenAI launches ChatGPT for Teens
- EdTech Innovation Hub: Common Sense Media to test ChatGPT for Teens
- Common Sense Media: ChatGPT and Sora pose risks to teens
- Common Sense Media: Launches Youth AI Safety Institute
- Benton Institute: Major AI chatbots unsafe for teen mental health support
The source trail.
Sources (9)
ChatGPT for Teens is an ‘unacceptable risk,’ says Common Sense Media
Article text retrieved; extracted text may omit tables or interactive elements.
theverge.com