AI Safety
How should schools respond when AI gives a dangerous or inappropriate answer?
Treat it as an incident: review the thread, support the student, notify the right adults, document what happened, and decide whether a setting, tool, or expectation needs to change. Do not leave it as a private student problem.
A dangerous or inappropriate answer is a school incident, not only a model failure. Response usually includes: stop the session if needed, review what was asked and what was returned, check on the student, involve counseling or administration when the content warrants it, and tell families when your policy says you will.
Then look at the system: Was a tool on that should have been off? Did filtering fail? Does the approved-tool list still make sense? Safety monitoring is only useful if someone owns the follow-up.