Claims and verification

AxioRank makes a small number of precise competitive claims. Each is stated here in full, with what it rests on and how you can check it yourself. We re-verify these against the shipping market every quarter; the dates below say when each was last checked. If you find one that no longer holds, tell us and we will fix it.

The transparency-log gateway

The only commercial AI agent gateway whose audit log is itself a hosted RFC 6962 Merkle transparency log: hourly Ed25519 signed tree heads with a public key endpoint, checkpoint APIs for independent log monitoring, and a per-action signed receipt for every governed call, each verifiable offline with a zero-dependency open source verifier in TypeScript and a CLI, plus a Python verifier that needs only the cryptography library.

Verify it: fetch the public key at /api/v1/audit/public-key, pull the checkpoint series from /api/v1/audit/log/{logId}/checkpoints, and verify a receipt offline with @axiorank/audit-verify in Node, Python, the CLI, or a browser build. You can run the browser build yourself at axiorank.com/verify/receipt, where the check happens locally and nothing you paste is uploaded. See Verify our log yourself.

Runtime integrity information-flow control

The first runtime integrity information-flow control for AI agents shipped in a commercial gateway, with no agent rewrite, no framework migration, and no special hardware: untrusted tool outputs are minted as value-level taint with token-level fingerprints, propagated across the trace, and enforced at egress and destructive sinks with deny, hold, or allow tiers.

The load-bearing words are integrity (where untrusted data came from, not just where sensitive data goes) and no agent rewrite (it is a gateway you put in front of agents you already run). Confidentiality-direction taint DLP exists elsewhere; value-level integrity taint on untrusted tool outputs, in a gateway, does not.

Prevention and proof in one receipt

The only AI agent gateway that signs all three into one offline-verifiable receipt: the information-flow policy the call was evaluated under, the enforcement decision itself, and the cryptographic signature of the human who approved any exception, bound into the delegation chain. Verifiable with nothing but the receipt and a public key: no TEE hardware, no ledger, no vendor in the loop.

The uniqueness is the three-way conjunction in one artifact. Verification needs two inputs you control: the receipt, and the published log public key you pin once.

Opt-in cross-tenant threat intelligence

The only AI agent gateway with an opt-in, customer-contributed cross-tenant threat intelligence exchange for agent and MCP tool-call attacks, protected by a published k of 5 anonymity floor: an indicator detected in one participating tenant enriches verification for every participating tenant, without exposing who contributed it.

The anonymity floor is published and enforced in code: an external identity is only ever surfaced once at least five distinct workspaces have independently flagged it. See Detection intelligence.

The public protocol coverage tracker

The only agent security control plane that publishes a protocol coverage tracker, across agent, auth, and payment protocols, with a public per-protocol status and live-in-the-gateway count.

Verify it: the live, machine-readable matrix is at /api/v1/protocols. The "live in the gateway" rows have a working adapter today; "watching" and "planned" rows are tracked but clearly marked as not yet live.

Conformance proof for AI-agent purchases

The first commercial AI agent gateway, as of July 2026, to enforce a user's signed spending mandate at the agent's tool-call boundary, before money moves, and emit a Commerce Conformance Seal a merchant, an issuer, and the buyer can each verify offline: an Ed25519-signed proof that binds the amount, currency, merchant, cart hash, the mandate hash, and an in-scope verdict, sealed into the same RFC 6962 transparency log with independent witness co-signing.

The load-bearing words are enforce at the tool-call boundary (the mandate is checked before the purchase call executes, not reconciled afterward) and offline-verifiable (any of the three parties checks the seal against a pinned key, and, holding the mandate, re-runs it to confirm the signed in-scope verdict is honest).

Verify it: verify a seal with verifyCommerceSeal in @axiorank/audit-verify, or at axiorank.com/verify/receipt. The split-view defense is the log's witness co-signatures (Sigstore Rekor and self-hostable signed-note witnesses), which a reader confirms with countValidWitnesses over the published checkpoints; a deployment's active witnesses are listed on each checkpoint. See Agent Commerce Conformance.

What this claim does not say: we do not claim a first on the mandate itself (Google AP2, Mastercard Agent Pay, and Visa Intelligent Commerce issue verifiable mandates), on agent identity (Visa Trusted Agent and others), or on checkout (OpenAI, Stripe, and the card networks). Those solve the front half. Our claim is scoped to proving the resulting purchase stayed inside the mandate, in a log that cannot equivocate. It is complementary to the mandate issuers, not a substitute.

Things we deliberately do not claim

• We do not claim "the first signed receipts for agents" (the receipts primitive is standardizing across several vendors and IETF drafts).
• We do not claim "the only Merkle audit log" (a Merkle logging service exists).
• We do not claim "the first offline-verifiable evidence bundle" (others ship one).
• We do not claim conformity with the EU AI Act; our pack is evidence supporting a deployer's obligations. See EU AI Act evidence pack.

Last re-verified: July 2026. Next review: October 2026.