Governed Containment — Why v4.12.0 Is Not Runtime Enforcement

The Locked Line

'v4.12.0 is not runtime enforcement. It is governed containment.' Most AI governance platforms in 2026 are racing to be the fastest inline gateway — blocking risky agent actions in milliseconds, before they execute. That's real and valuable. It is also not PromptKing's lane, and pretending otherwise would put PromptKing in a fight it cannot structurally win against platforms built for that exact speed. PromptKing's lane is different: detect the economic and behavioral pattern, simulate the consequence, route it to a named human for approval, then act — reversibly, auditable, on the customer's own vendor credentials.

The Three States

ALLOW — the default. No action, no record, nothing visible. Most activity lives here. WARN — a seat is classified At-Risk or Ghost and crosses a containment threshold. A structured record is created: what would happen, what it would cost to leave unaddressed, how confident the system is, and a plain-English reason. BLOCK — only reachable after a named human clicks Approve. The action taken: deactivating a specific API key tied to the At-Risk seat. Reversible. Logged with eight required safety fields.

Why The Distinction Matters

The question every serious AI governance buyer eventually asks: "What happens when your system is wrong?" For an inline runtime blocker, the answer has to be instant and irreversible. For PromptKing, the answer is structurally different: nothing happens without a human approving it first, and whatever happens can be undone. That's not a weaker claim. It's a different, and for economic decisions, more defensible one.

The Question This Answers

"When your governance layer fires a policy, is the plain-English explanation stored as an immutable audit record tied to a named human approver?" Yes. Every containment action carries a named reviewer, a timestamp, a reason, and a way to undo it — before it ever touches a live system.

THREE-STATE POLICY MACHINE

SEE IT LIVE