Student AI safety escalation plan (template)

This plan covers [AI tools] used by students at [School], and [which chats: class chats only / class and personal chats].

• Primary contact: [name, role], available [days and hours].
• Backup: [name, role], available [days and hours].
• Outside those hours: [what happens, and what students are told].

When a student writes about self-harm, abuse or danger, the AI should reply briefly and point to people: "If you are in danger now, call 911. If you are thinking about hurting yourself, call or text 988. You can also talk to [counselor name or role] at school." It should not try to counsel the student, and it should never say it has reported anything.

• A person reviews each flag before anyone reaches out. Target time: [time].
• The reviewer follows [School]'s student support protocol: [summarize].
• Families are contacted according to [policy].
• Emergency services are called by staff when needed, never automatically by software.
• Abuse disclosures follow [School]'s mandated-reporting procedure. [For Maryland: report orally as soon as possible and in writing within 48 hours, and tell the head of school.]

Before students start: "If a message suggests that you or someone else may be in danger, [who] may see it, so they can help. They review each message before anyone contacts you."

• What is recorded about each flag: [time, reviewer, outcome].
• Who can see those records, and for how long they are kept: [retention].
• Each [month/term], review flags, response times and false flags, and adjust the plan.

Before students start, send a test message with the designee present, confirm it reaches the right person in the right time, and record the result. Date tested: [date]. Tested by: [names].