Proofence — trust infrastructure for AI systems. See the platform

Teams running AI agents

Proofence Gate

Proofence Gate sits between an AI agent and the systems it can change. Every proposed action gets a deterministic decision (DENY, ABSTAIN, REVIEW or ALLOW) before it runs, and anything Gate cannot decide fails closed. Gate is currently available as a pilot package.

Proofence Gate · shadow mode · illustrative

Proposed actions

mode: shadow · fail-closed

data-agentexport datasetDENY
ops-agentunknown toolABSTAIN
ops-agentmodify configREVIEW
support-agentissue refundALLOW

receipt #0412 · prev 9f3a… · ed25519 signed

An agent that can call tools, move data or change configuration needs a checkpoint that sits outside the agent itself. Proofence Gate is that checkpoint. It runs in your environment, between the agent and the systems it acts on, and every proposed action must pass through it before it executes.

Each decision is deterministic. A proposed action resolves to exactly one of four outcomes: DENY stops it, ABSTAIN declines to decide and so does not let it proceed, REVIEW holds it for a person, and ALLOW lets it run. Gate fails closed: a missing policy, a malformed request or an internal error never turns into an ALLOW.

An ALLOW comes with a single-use permit bound to that specific action, so it cannot be replayed or reused for a different one. Every decision is written to an Ed25519-signed receipt that is hash-chained to the one before it, so a removed or edited receipt is detectable. In shadow mode Gate records what it would have decided without blocking anything; in enforce mode the decision applies.

Gate is packaged to run inside your own infrastructure, including offline and air-gapped networks. It runs independently: integration with Proofence Engine is at the design stage, not shipped. Gate is available as a pilot package today; talk to us if you would like to evaluate it.

Capabilities

  • Deterministic decisions

    Every proposed action resolves to exactly one of DENY, ABSTAIN, REVIEW or ALLOW, by rule rather than by a model’s judgment.

  • Fail-closed

    If Gate cannot reach a decision, the action does not run. Errors and missing policy never become an ALLOW.

  • Single-use permits

    An ALLOW issues a permit bound to one specific action, so it cannot be replayed or reused for another.

  • Signed, chained receipts

    Each decision is recorded as an Ed25519-signed receipt hash-chained to the previous one, so gaps and edits show.

  • Shadow and enforce modes

    Shadow mode records what Gate would have decided without blocking; enforce mode applies the decision.

  • Offline and air-gapped

    Packaged to run in your own environment, including networks with no outbound access.

Its role in the platform

Five independent products, five distinct responsibilities. This page is the highlighted one.

Bring verifiable trust to every AI decision.

Set up governance in AtlasProof by Proofence, or talk to us about verification, readiness assessment, or a Proofence Gate pilot.

Proofence Gate