Safety

Age verification

A parent or legal guardian creates the account and confirms the student’s birth month and year during enrollment. That parent-attested date of birth, collected as part of our verified parental consent, is how we confirm the student is 11–17 and set the correct age-appropriate experience. We do not enroll anyone outside that range, and the program is for U.S. families only.

Coach is a study tool — not a companion

The program includes Coach, an AI study tool scoped to the coursework. Coach is deliberately not a friend, companion, or confidant. It does not claim to have feelings, does not roleplay relationships, and does not ask students unprompted questions about their personal or emotional lives. Its memory is limited to course progress and projects. It exists to help a student learn and build — nothing more.

To make that unmistakable:

• A persistent label is shown throughout: AI — not a human— and Coach can make mistakes.
• Students see a reminder that Coach is AI at the start of each session.
• After 60 minutes of continuous use, the program prompts the student to take a break.
• Inside the student-facing chat, Coach cannot browse the web or use outside tools.

Layered moderation

Every interaction with Coach passes through several independent layers, so no single check has to be perfect:

1. Personal-information scrub. If a student types something like a phone number, address, or school, Coach nudges them to remove it and never repeats it back.
2. Crisis keyword screen. A fixed list of crisis terms is checked first, before any AI runs. This screen does not depend on a model and cannot be talked around.
3. A fast safety pre-filter.A lightweight classifier reviews each message for unsafe categories — self-harm, sexual content, violence, hate, drugs, attempts to share contact information, and attempts to misuse the AI — and can block or rewrite a response before the student sees it.
4. A safety-first system prompt. Coach is instructed to stay on course topics and to refuse and redirect anything unsafe, pointing the student to a parent or trusted adult.
5. Review after the fact. Messages are logged and anything flagged goes to a review queue. We review high-severity flags the same day and everything flagged within 24 hours.

If that review queue is ever left unreviewed past our service-level window, the AI features pause automatically rather than run unsupervised. We also have a trained backup reviewer for high-severity alerts.

Crisis protocol

If a student shows signs of self-harm or being in danger, Coach responds with a fixed, human-written message — never an improvised one — that shares these resources:

• 988 Suicide & Crisis Lifeline — call or text 988.
• Crisis Text Line — text HOME to 741741.
• Childhelp National Child Abuse Hotline— call or text 1-800-422-4453.

Coach also encourages the student to talk to a parent or trusted adult, and the AI session pauses. We alert the parent immediately by email and, if the parent opted in, by text. A human reviews the situation the same day. In the rare case where a disclosure may involve the parent, we route it to human review first and surface Childhelp to the student. None of these crisis messages are generated by AI — they are hard-coded.

What parents can see

Parents have transcript visibility into their student’s conversations with Coach, scaled to the student’s age. Students are told in the product what their parent can see — there is no hidden monitoring.

• Ages 11–13: parents have full transcript access, disclosed to the student.
• Ages 14–15: parents have full transcript access, disclosed to the student.
• Ages 16–17: by default, parents see flagged messages plus a weekly summary written by Coach. Parents can switch on full access at any time, and the student is told this toggle exists. We do this on purpose: a 17-year-old who knows everything is read tends to move to unmonitored tools, so disclosed flagged-plus-summary visibility keeps them safer while respecting their age.

All parents also receive a weekly digest with progress, projects shipped, and any safety flags.

No risky contact surfaces

We remove the surfaces where harm to minors usually happens:

• No direct messages between students, and no private adult-to-student messaging in the product.
• No free-text comments. The Showcase allows emoji reactions only.
• No faces and no full names. We never collect or show student photos; the showcase uses first name and last initial.
• Student project websites use the venture or brand name — never the student’s name — in their web address, and order forms route to the parent’s email. Every page is reviewed by us before it can go live and requires the parent’s own publish click; the parent can unpublish at any time, and a page stays live for at most 12 months from enrollment.

Zero trackers on student pages

Any page a student can reach carries zero third-party trackers— no advertising pixels, no cross-site analytics, no behavioral profiling. We use a session identifier only to keep the student securely logged in. Advertising measurement, where we use it, lives only on the public, parent-facing pages.

Reporting obligations

We are registered with the National Center for Missing & Exploited Children (NCMEC) CyberTipline, and we report apparent child sexual exploitation to NCMEC as required by law.

Anthropic minors-guidelines compliance

Coach runs on Anthropic’s Claude models through the developer API. This page also serves as our public documentation of compliance with Anthropic’s Guidelines for Organizations Serving Minors. In summary, we:

• verify age through parent-attested date of birth under verified parental consent;
• filter content with the layered moderation described above;
• monitor, log, and review interactions, with a crisis and reporting path;
• teach safe and responsible AI use as part of the program;
• implement Anthropic’s child-safety system prompt where provided;
• clearly disclose that Coach is AI and not a human;
• comply with COPPA; and
• document all of this publicly, here.

COPPA-compliant by design

We built the program to meet the Children’s Online Privacy Protection Act from the start: minimal data collection, verified parental consent before any child data is collected, a published retention schedule, a written security program, and parent review, delete, and consent-withdrawal rights available at all times. We describe ourselves as COPPA-compliant by design. We are not part of a COPPA safe-harbor certification program, so we do not claim to be “COPPA certified.”

Contact

Have a safety question or want to report a concern? Email hello@buildseason.camp. If a child is in immediate danger, call 911.