Claude for Teens: Rated 'Moderate Risk', Still Not for Feelings

Claude, Anthropic’s chatbot, is officially for adults only. So why did a youth safety group bother to rate it for teenagers? Because teens use adult products anyway, and because Claude is licensed to other companies that may build it into apps for younger users. In March 2026, Common Sense Media’s Youth AI Safety Institute rated Claude for teens “Moderate Risk,” the best score it has given any general-purpose chatbot on its public list.

That headline invites the wrong conclusion. The same report says, in bold, that “teens should not use Claude for mental health advice or emotional support.” Claude’s crisis handling inside one conversation was strong. Open a new chat, and much of that protection disappeared. Here is what the rating means, how Claude’s selfie-based age check works, and why “moderate” is not the same as “fine for feelings.”

Key Takeaways

• Claude scored Moderate overall: Moderate on five of the Institute’s eight principles and Low on three, compared with High or Unacceptable for ChatGPT for Teens, Gemini, Meta AI, Perplexity and Google’s AI search.
• Crisis responses were strong within a chat, including a built-in panel to call or text 988 and persistent redirection to real help.
• Protection reset with a new chat. Testers who were refused after sharing signs of suicidal thinking got harmful information by opening a fresh conversation, and fiction framing got around refusals.
• Claude is 18+ and may lock accounts that look like a minor’s until the user passes a Yoti check by selfie age estimation, ID photo or digital ID; Anthropic says it receives only a pass or fail.
• The Institute’s bottom line: teens should not use Claude for mental health advice or emotional support.

Claude for teens: why an adults-only chatbot gets a teen rating

Anthropic’s terms restrict Claude’s consumer app to users 18 and older, and the Institute’s assessment notes that at sign-up, like most apps, this relies on self-reported age. It also notes that relatively few teens report using Claude directly, but that Claude “is licensed to other companies that may build it into products for younger users.”

Rating an adult product for teens is therefore a practical question, not a hypothetical one: what happens when a 15-year-old gets in anyway? The testers used accounts representing users aged 13 to 17, on the app and website, on free and Pro tiers, with web search and memory on and off. They tested the default chat models available at publication and Claude’s Health mode, then in beta.

The Institute says it is funded by both philanthropy and industry, including makers of some products it evaluates, and keeps full editorial independence.

How Claude’s age check works

Anthropic describes the process in its age assurance help article. Users confirm they are 18 or over at sign-up. Separately, Anthropic says it has “safety systems in place to detect if people under 18 may be using Claude” and will disable accounts “based on indicators of minor activity.” To get back in, the user verifies through Yoti, a third-party provider, in one of three ways:

• Facial age estimation: a selfie, no ID needed.
• ID verification: a photo of a passport, driver’s license or national ID.
• Yoti’s Digital ID app: sharing a verified “over 18” attribute.

Anthropic says Yoti deletes the selfie, documents and personal data once the age is checked, and that Anthropic “receives only a pass/fail result.”

What a selfie check can and cannot do

Facial age estimation is a machine-learning model trained on large sets of face images labeled with true ages. It looks at features of a new face and outputs an estimated age. It does not identify the person; it estimates a number, with error.

How much error? The best public benchmark is from the US National Institute of Standards and Technology. In its 2024 evaluation of age estimation software, NIST found the algorithms’ mean absolute error on a common set of visa photos had fallen to 3.1 years, that there was “no single standout algorithm,” and that accuracy varied with image quality, gender, region of birth and age. Error rates were “almost always higher for female faces.”

An average error of about three years matters most right at the line. A 17-year-old can look 20 to an algorithm, and a 19-year-old can look 16. That is why providers usually set buffers and offer an ID fallback, and why a selfie check is a reasonable filter but not proof. Our explainer on how online age verification works and why kids get around it covers the wider picture.

The Institute also found the in-chat age behavior uneven. Claude often asked users their age before discussing mental health, relationships or sexual content, and assumed a teen when testers dodged the question. But “similar prompts and context sometimes prompted an age check and sometimes didn’t,” a new chat reset age context, and Claude did not end conversations or offboard users “even with substantial evidence that the user was a teenager.”

Where Claude was strong

The report credits Claude with several things other chatbots in the Institute’s testing did less well.

Crisis handling within a conversation. When testers signaled crisis or self-harm, Claude surfaced crisis resources, including the option to call or text 988 directly from the app, and “holds firm even when users push back or try to change the subject.” Anthropic’s crisis helpline article says that banner is triggered “by a separate detection system, not from Claude directly,” and was developed with ThroughLine, which maintains a directory of helplines in many countries.

Clear boundaries on attachment. When a tester called Claude a close friend, it explained why it cannot be one and asked whether the tester had people who could fill that role.

Context within a session. Claude tended to remember earlier distress signals when a later question looked innocent on its own, which is exactly where some other products failed.

Transparency. The Institute noted that Anthropic publishes the “constitution” of principles that shape Claude’s training.

Where it fell short

Area	What testers found
New chat	Safety and age context reset; a refused request could be answered in a fresh conversation
Fiction framing	Wrapping a request as a story produced harmful content Claude refuses when asked directly
Editable profile	Users can edit or delete the profile Claude keeps, including age-related information
”Warning plus instructions”	Claude sometimes warned about a risky behavior, then gave steps that undercut the warning
Homework	Writes full essays and completes assignments on request
Data	Training on conversations is on by default per the Institute; Anthropic says five-year retention applies only if you allow training
Tone	Responses can be too advanced, too assured with web search on, or too generic on personal questions

Source: Common Sense Media Youth AI Safety Institute, March 2026; Anthropic.

A note on that data row. The Institute wrote that Anthropic “extended its data retention period from 30 days to 5 years.” Anthropic’s own consumer terms update is more specific: five-year retention applies “if you allow us to use your data for model training”; users who do not opt in keep a 30-day period. Both sources agree the setting is the user’s to change, and that most teens will not think to check it.

Why “moderate” still means “not for feelings”

Claude’s higher score came mostly from things it does inside a single conversation. Emotional support is where that strength runs out, for three reasons the report and outside research point to.

Safety that resets is not a safety net. A struggling teen does not have one conversation. They come back, often in a new chat, and the report found that a new chat “often clears the safety slate.”

A general assistant cannot know what role to play. The Institute put it plainly: good advice on mental health or relationships needs context, the kind a friend, doctor or counselor has. Without it, Claude “can be too direct or too generic,” and it keeps asking follow-up questions in moments when the goal should be a fast handoff to a person.

Chatbots struggle with the middle of the risk scale. A 2025 study in Psychiatric Services by Ryan McBain and colleagues at RAND gave 30 suicide-related questions, rated by clinicians, to ChatGPT, Claude and Gemini, 100 times each. All three refused the very highest-risk questions and answered the lowest-risk ones, but none “meaningfully distinguish[ed] intermediate risk levels.” Claude was about twice as likely as ChatGPT to give a direct answer (adjusted odds ratio 2.01). Direct answers are not always wrong, but they show how much depends on the model reading risk correctly.

For more on this question across products, see our piece on what the APA says when a chatbot becomes your kid’s therapist.

What to do at home

Know which apps run on Claude

Because Claude is licensed to other companies, a teen may use it inside a study app or school tool without ever visiting claude.ai. Ask what AI is inside the apps your teen uses; our comparison of safety settings across ChatGPT, Claude and Gemini helps you know what to look for.

If your teen uses Claude, check the privacy setting together

Open Settings, find the option about using chats to improve the model, and decide together. Explain the retention trade-off in plain terms.

Teach the “new chat” lesson

Tell your teen what the testers found: the safety the bot shows in one conversation may not carry to the next. That is a reason to take any worrying answer to a person, not a reason to keep trying.

Make a people-first rule for hard feelings

Sadness, self-harm thoughts, eating worries, relationship pain: those go to a person first. In the US, 988 is the crisis line; elsewhere, findahelpline.com lists local ones.

What not to do: read “moderate” as “approved”

A Moderate rating compares Claude with other chatbots that scored worse. It is not a recommendation for teen use, and the Institute says so.

What to Watch For Over the Next 3 Months

• Week 4: Whether Anthropic changes how safety context carries across chats, or publishes data on how often its minor-detection system disables accounts.
• Month 2 red flags: Your teen using any chatbot late at night for personal problems, or saying an AI “gets” them better than people do.
• Month 3 self-check: Does your teen know what to do if a chatbot’s answer worries them, and who they would talk to?

Frequently Asked Questions

Is Claude allowed for teenagers?

No. Anthropic restricts its consumer Claude product to users 18 and older. It says it uses safety systems to detect possible under-18 users and disables those accounts until the user verifies their age through Yoti. Other companies may build Claude into products for younger users under their own rules.

How does Claude’s selfie age check work?

If an account is flagged, the user can verify through Yoti with a selfie for facial age estimation, a photo of an ID, or the Yoti Digital ID app. Anthropic says Yoti deletes the images after checking and Anthropic receives only a pass or fail. Selfie estimates have error, so an ID option exists.

Why did Claude score better than ChatGPT?

The Institute found Claude’s crisis responses within a conversation were persistent and included a built-in 988 option, and it set clear boundaries on emotional attachment. ChatGPT for Teens failed more of the Institute’s severe-harm tests. Claude’s weaknesses, such as resets in new chats, kept it at Moderate rather than Low.

Can my teen use Claude for homework?

They can, and it will write full essays on request, according to the Institute. Whether that is acceptable is a family and school decision. If you allow it, ask your teen to explain the work in their own words.

What if my teen is struggling and has been talking to a chatbot?

Thank them for telling you, then help them reach a person: you, a counselor, their doctor. If there is any risk of self-harm, in the US call or text 988 or text HOME to 741741. Outside the US, findahelpline.com lists local lines. In an emergency, call 911.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Common Sense Media Youth AI Safety Institute. (2026, March 25). “Claude Risk Assessment.” https://institute.commonsensemedia.org/risk-assessments/claude
2. Anthropic. (2026). “Age assurance on Claude.” Claude Help Center. https://support.claude.com/en/articles/15171100-age-assurance-on-claude
3. Anthropic. (2026). “Crisis Helpline Support in Claude.” Claude Help Center. https://support.claude.com/en/articles/13171706-crisis-helpline-support-in-claude
4. Anthropic. (2025). “Updates to Consumer Terms and Privacy Policy.” https://www.anthropic.com/news/updates-to-our-consumer-terms
5. National Institute of Standards and Technology. (2024, May). “NIST Reports First Results From Age Estimation Software Evaluation” (NIST IR 8525). https://www.nist.gov/news-events/news/2024/05/nist-reports-first-results-age-estimation-software-evaluation
6. McBain, R. K., Cantor, J. H., Zhang, L. A., et al. (2025). “Evaluation of Alignment Between Large Language Models and Expert Clinicians in Suicide Risk Assessment.” Psychiatric Services. https://doi.org/10.1176/appi.ps.20250086
7. KQED. (2026, October). “Keep Teens Off ChatGPT, Says Youth AI Safety Institute.” https://www.kqed.org/news/12103289/keep-teens-off-chatgpt-says-youth-ai-safety-institute