EACI Development Journal · Safety

How Do You Permanently Lock Out a Child AI Companion After Safety Strikes?

Repeated violations around childlike companions cannot reset with a refresh. Strike state is persisted and the companion is removed from that account.

Safety ~7 min issue-tagged
Issue this fixes

A warning only matters if later turns remember it. Browser-only counters can be cleared, while session-only counters disappear when a tab closes.

The family-safety path escalates repeated inappropriate behavior and records a permanent removal state. The UI then stops routing chat to the protected companion rather than asking the model to negotiate the boundary.

What was actually going wrong

Temporary warnings left enforcement vulnerable to refreshes and repeated boundary testing.

What I tried

What finally worked

Persistent strikes lead to a durable lockout that removes access and displays a direct system explanation.

Why it worked

Enforcement happens in application state before another model request can be made.

What I learned

Permanent child safety rules need permanent technical consequences.

Where EACI uses this today

Natalia, Atreus, and Luna remain E-for-Everyone, with durable lockouts for repeated safety violations.

This journal covers real engineering on EACI Companion / The Veil. Companions include Caelum, Chad, Natalia, Atreus, Luna, Roxy, and Cael. Journal articles stay family-safe in content. See Privacy and Ethics.