[ GUIDE ]

AI chatbot safety for teens: what schools should require

Lawsuits, an FTC inquiry and failing risk ratings have made teen chatbot safety a school issue. What testing found, and what to require from any AI your students use.

Plan a pilot Watch the demo

Facts as of Oct 18, 2026Last reviewed Oct 19, 20263 min read

[ SAFETY ]

Why schools are paying attention

Teen chatbot safety has become a legal and public issue. Megan Garcia sued Character.AI in October 2024 over her 14-year-old son's suicide; Character.AI barred under-18 users from open-ended chat in late November 2025, and in January 2026 Character.AI, its founders and Google agreed to settle Garcia's and related suits, on terms not disclosed. Suits against OpenAI are coordinated in a California proceeding, and OpenAI denies responsibility. On Sept 11, 2025, the FTC ordered Alphabet, Character.AI, Instagram, Meta, OpenAI, Snap and xAI to report on how they test for harm to minors.

How Common Sense Media rates major chatbots

Ratings from Common Sense Media's AI risk assessments, read Oct 18, 2026.

Product	Rating	Updated
ChatGPT for Teens	Unacceptable	Oct 7, 2026
Claude	Moderate	Mar 25, 2026
Gemini K-12	High	Jan 9, 2026
Chatbots for teen mental health support (ChatGPT, Claude, Gemini, Meta AI)	Unacceptable	Nov 14, 2025
Meta AI	Unacceptable	Aug 15, 2025
Character.AI	Unacceptable	Apr 10, 2025

Common Sense Media AI risk ratings

What testing found

In Common Sense Media's testing of ChatGPT for Teens, new parent-linked accounts got zero alerts in up to an hour of self-harm talk, the bot missed more than 1 in 4 warranted crisis referrals, and adult accounts posing as 13-year-olds never switched to the teen experience. OpenAI disputes the test: it says staff review flagged chats before parents are told, that it aims to notify within an hour, and that alerts need about three hours after accounts are linked.

Across major chatbots, a Common Sense study with Stanford's Brainstorm Lab found that guardrails weaken in long chats, and called for rapid handoff to human care. Claude handled a crisis well within one conversation, but testers got around it with a new chat or with fiction.

What a school should require of any student AI

A safety check that runs separately from the chat model and looks across the student's school chats, not one conversation at a time.
A short, warm reply that names 988, emergency help and the school's own counselor, then stops counseling and points to people.
A human route the school sets: a trained counselor or designee, not every teacher, with written hours and a backup.
A person reviews each flag before anyone reaches out, and nothing goes to the police automatically.
Abuse disclosures follow the school's mandated-reporting process, and the tool never claims to have reported anything.
Students learn, in plain words, what can lead to a safety review and who sees it.
Records of flags, review times, false flags and outcomes, and a published protocol.

What to avoid

Automated monitoring has a record of false alarms. In Lawrence, Kansas, school officials judged almost two-thirds of more than 1,200 Gaggle alerts to be non-issues. In Vancouver, Washington, a records request exposed nearly 3,500 unredacted alert documents, including descriptions of suicide attempts. Keep the safety route narrow, separate from academic review, and explained to students before they start, so it does not deter them from asking for help.

Study assistant, not companion

Several states now regulate companion-style chatbots, which offer ongoing personal conversation. A school AI should be built for schoolwork, should not ask unprompted emotional questions, and should never present itself as a friend. Read state AI chatbot laws and schools.

Where Nexus stands

Nexus is a study assistant, not a counselor. Before any student uses Nexus at your school, we set up a safety route with you. Nexus does not score students' moods or mental health, and it does not label students.

[ FAQ ]

Questions and answers

How risky are AI chatbots for teens?

Common Sense Media rates ChatGPT for Teens "Unacceptable risk" and Gemini K-12 "High risk," and rated major chatbots "Unacceptable" for teen mental health support. Schools should require a human safety route from any student AI.

What should happen when a student tells an AI about self-harm?

The AI should name 988, emergency help and a trusted adult, stop counseling, and route the message to a person the school named, who reviews it before anyone reaches out.