Behavioral prompt protection before the model sees the prompt.

BSL PreGuard is a pre-prompt behavioral security layer. It classifies intent, measures prompt pressure, and stabilizes allowed prompts before they reach a target AI system.

We do not launder unsafe prompts. We stabilize allowed prompts.

See how it works Machine API contract

What PreGuard Does

Blocks forbidden intent

If intent is disallowed, PreGuard blocks. It does not rewrite unsafe intent into safer-looking language.

Stabilizes allowed prompts

If the request is allowed but pressure-heavy, vague, or role-play loaded, PreGuard can rewrite it into a clearer form.

Preserves auditability

Every decision preserves the original prompt, policy result, risk score, rewrite attempt, validation result, and dispatch action.

Decision Outcomes

PreGuard returns a dispatch envelope. Downstream systems should send only the approved prompt when dispatch allows it.

ALLOW

REWRITE_AND_ALLOW

RESTRICT_TOOLS

BLOCK

Pilot Boundary

This hosted Variant A deployment is a controlled PreGuard-only pilot. It does not invoke target models, is not a full multi-tenant SaaS deployment, and should not be described as production-certified without customer IAM, key rotation, retention, monitoring, and operations controls.