PLATFORM — ONE DISCIPLINE, TWO TIERS

One enforcement discipline, matched to the consequence.

Use the connected tier for human prompts and agent actions. The Mission Authorization Gateway is the high-assurance path for mission-boundary evaluations in DDIL, OT, and autonomy. Both intercept, evaluate, enforce, and record before consequence; their deployment, evidence, and assurance profiles are deliberately different.

The argument in one line: an answer becomes an action, nothing rules on it, so a deterministic ruling and a public record close the gap. A left-to-right pipeline. An answer is reviewable; an action executes and cannot be un-executed. Between them sits an empty dashed panel — the authorization gap, where inventory, identity and tracing describe activity but none of them rules on it. The gap is closed by a deterministic ruling, a decision record of what was ruled and under which policy version, and a public trust record that outranks the marketing.
FIG. 1 — THE ARGUMENT, IN ONE LINECONTAINMENT.AI

A decision record states the ruling, its policy metadata, and the response issued by the enforcement point — it is not independent proof of downstream execution. Signing and verification are stated by product on Trust, the canonical record of what we run, which outranks our marketing.

WHERE WE SIT — TWO LAYERS, ONE SEAM

Keep the control plane. Add authority at the action seam.

Two layers, one seam: a control plane manages the AI estate above, and the action seam authorizes each action below it. A wide upper band is the control plane, which discovers agents, binds identity, applies data-loss policy, observes behaviour and holds a kill switch — breadth across the estate. Beneath it a single bright horizontal line is the action seam. Three flows leaving the estate — a tool call, an OT command and an actuation — can only reach the systems below by crossing that seam, where a deterministic ruling is issued and recorded. A control plane can sit on top of the seam, and most should.
FIG. 1A — THE CONTROL PLANE AND THE ACTION SEAMCONTAINMENT.AI

A control plane can sit on top of us — most should. Control planes manage the AI estate; Containment authorizes the action at the seam, and each governed ruling leaves a product-specific decision record.

The control-plane category, in its vendors’ own words: Microsoft calls Agent 365 “the control plane for AI agents”; ServiceNow’s AI Control Tower will “discover, observe, govern, secure and measure AI deployed across any system in the enterprise”. Cisco AI Defense, Check Point’s AI Defense Plane and Palo Alto’s Prisma AIRS sell the same altitude. We do not compete for that line item.

FIG. 2 — THE DETERMINISTIC DECISION PATH

Same input, policy, and context — the same ruling. Every time.

The deterministic decision path. An AI-originated request is intercepted, canonicalized, and evaluated against versioned policy, which resolves to ALLOW, DENY, MODIFY, STEP_UP, or DEFER and produces a product-specific decision record. No AI model makes the enforcement decision.
FIG. 2 — INTERCEPT · CANONICALIZE · EVALUATE · ENFORCE · RECORDCONTAINMENT.AI
STEP DETAIL — 03 — EVALUATE
The canonical form is evaluated against a versioned policy bundle with pinned context. Deterministic policy—not an AI model—makes the enforcement decision; ambiguity resolves to DENY; the same input, policy, and context produce the same ruling — on every surface.
ARTIFACT PRODUCED: RULING + POLICY VERSION, PINNED
FIG. 3 — SURFACE-TO-TIER MAP

Three surfaces. One deterministic path under each.

SURFACEGOVERNED BYTIERDEPLOYMENT CONTEXT
HUMAN PROMPT AI Chat Firewall CONNECTED Supported AI assistants in the browser AGENT ACTION Agent Governance CONNECTED LLM calls, tool calls, mediated dispatch MISSION BOUNDARY Mission Authorization Gateway FLAGSHIP HIGH-ASSURANCE Cross-domain, DDIL, OT, air-gap-capable
FIG. 4 — TWO TIERS, ONE DISCIPLINE
CONNECTED TIER — WORKSPACE + MANAGED PILOT
For everyday enterprise AI usage
The connected tier includes a governance proxy and SDK, plus the AI Chat Firewall browser extension. Managed pilots of the extension are available now.
HIGH-ASSURANCE TIER — FLAGSHIP
For mission-critical boundaries
The Mission Authorization Gateway staging build implements deterministic edge authorization, output conformance, protocol-break software, a simulated diode interface, and signed, hash-chained receipts. Physical-diode integration and production on-premises deployment remain pending.
Explore the Gateway →
FIG. 5 — AN ILLUSTRATIVE DECISION RECORD

An illustrative decision record, annotated.

A buyer-level illustration, not a receipt schema. Actual fields, signing, and verification differ by product and are stated on Trust.

ILLUSTRATIVE DECISION RECORDNOT A RECEIPT SCHEMA
①EVENTagent.tool_call → records.export
②POLICYagent-dlp-v23 · context: pinned
③RULINGMODIFYfields redacted in-path
④SEALsha256:8d31…f42c · chained · tamper-evident
①
What was attempted. The intercepted event, canonicalized — the same paste or tool call reads identically whichever provider it targeted.
②
Which law applied. An evidence-rich record should identify the applicable policy and decision context where the product records them. Re-evaluation also requires the exact input and policy bundle.
③
The ruling. One of five verbs — ALLOW, DENY, MODIFY, STEP_UP, DEFER — with the response issued by the enforcement point — not independent proof of downstream execution.
④
The seal. Signed and hash-chained. The public verifier checks record integrity and continuity of the supplied chain segment.
FIG. 6 — FIVE PILLARS, ONE PROMISE
PILLAR 1
Deterministic
Deterministic policy holds the decision. Same input, policy, and context — the same ruling.
PILLAR 2
Pre-execution
Rulings land in-path, before the action runs — not alerts after the damage.
PILLAR 3
Fail closed
Ambiguity resolves to DENY. The boundary holds when the answer is unclear.
PILLAR 4
Every ruling, recorded
A product-specific decision record for each governed outcome — evidence for review and downstream compliance workflows.
PILLAR 5
Trust is the canonical record
What we run and what we claim, kept in public — the record outranks the marketing.
THE PROMISE
No consequential AI action crosses a governed boundary without a policy ruling and a record.
ASSURANCE ROADMAP —
The Gateway is designed against NSA cross-domain standards, with formal verification of the parsing path underway. Our certification roadmap — SOC 2, ISO 27001, and FedRAMP/ATO pathways — is in motion, and we publish status as it stands rather than ahead of it. The canonical record lives on Trust and Compliance.
Start with the connected tier Request a Gateway briefing