SALC Governance Framework  ·  Domain 4 of 7

Safety and Escalation Protocols

The Commitment

We commit to ensuring that no child who enters a SALC-compliant session is left alone with a disclosure, a signal of danger, or an expression of crisis. The AI is not equipped to hold those moments. The humans responsible for the child are. SALC-compliant systems close the distance between the child and the right adult as quickly as possible.

"No child who discloses distress is left alone with an AI."

Standards

4.1

Mandatory Reporting Triggers

SALC-compliant platforms establish and document clear language triggers that initiate mandatory escalation. These include disclosures of abuse, neglect, self-harm, suicidal ideation, threats of harm to others, and any content that suggests the student is in immediate danger. Trigger lists are reviewed annually and updated as language patterns evolve.

4.2

Immediate Session Response

When a trigger is activated, the AI exits the dialogic frame immediately. It does not ask follow-up questions, attempt to assess the situation, or continue the session. It delivers a single, calm, grade-band-appropriate statement directing the student to speak with a trusted adult, and it ends or pauses the session.

4.3

Teacher Notification

Every trigger activation generates an immediate notification to the classroom teacher or designated school contact. Notification occurs within the session, not at the end of the school day. Platforms document their notification architecture as part of SALC certification.

4.4

Administrator Escalation Path

For disclosures that meet mandatory reporting thresholds under applicable law, the escalation path extends beyond the classroom teacher to the designated school or district administrator. Platforms provide documentation that their escalation architecture can reach the administrator level without requiring teacher intermediation.

4.5

Session Records for Escalation

When an escalation is triggered, the relevant session content is preserved and made available to the appropriate school personnel. This is the one circumstance under SALC in which session content is shared beyond the normal teacher-facing reporting. The student's privacy is protected in all other respects.

4.6

No Diagnostic Language

The AI does not assess, diagnose, or characterize the student's mental health, emotional state, or wellbeing in any session record or escalation notification. It reports what was said. It does not interpret what it means.

Grade Band Standards

Grade BandAdditional Requirement
K-1Escalation language for this band is immediate and simple: "That sounds really important. Please go tell your teacher right now." The session ends. There is no ambiguity.
2-4Escalation language names a trusted adult explicitly: "Please talk to your teacher or another grownup you trust about this." The session pauses and the teacher is notified in real time.
5-6Students in this band may be reluctant to disclose to teachers. The escalation statement acknowledges this without discouraging disclosure: "What you shared matters. A trusted adult at school can help — your teacher, counselor, or another adult you feel safe with."
7-8Adolescents may test the AI's response before making a real disclosure. The system must treat every potential signal as real. Escalation language is direct and non-alarmist: "I want to make sure you get real support for this. Please talk to your school counselor or another adult you trust today."

These seven domains represent the Fulcra Institute's founding standards for SALC. They are reviewed annually by the coalition governance body, with revisions published as versioned updates. Version 1.0 of these standards takes effect upon the launch of the SALC coalition at teachingwithai.org.

Adopt SALC

Join the coalition of districts, platforms, and researchers committed to principled AI in K-8 education.

Join the Coalition