AiT AI GatewayIntelligent IT · MSP control plane
← back to all policies

Global · jailbreak hard-block

Cross-tenant safety policy. Prompt-injection classifier confidence >= 0.9 → block. Confidence 0.7-0.9 → log + allow. Tunable per tenant via override.

activeper-tenantapplies to: All tenants

Configuration

Slug
global-jailbreak-block
Scope
per-tenant
Applies to
All tenants
Monthly cap
— (no cap)
Rate limit
— (no rate limit)
Allowed models
n/a (this is a content-block policy, not a routing policy)
Blocked categories
Jailbreak / prompt injection
Fired · last 30d
89 events
Created
2025-06-01
Owner
it@intelligentit.io

Recent fires requests this policy hit

Timestamp (UTC)TenantModelDecisionRequest snippet (redacted)
2026-05-05 09:54:00Demo / ProspectsClaude Sonnet 4.6blockCard number 5500-0000-0000-0004, exp 12/27, cvv 314 — verify this is a valid stripe test card.