Verify the merchant before your agent pays.
One API call. Address, company, domain‑binding, sanctions — in the agent's loop.
Illustrative demo. Acuris verifies merchant legitimacy from official registers before an agent pays.
The one call your agent makes before it pays
Give us the storefront URL and who it claims to be — a name, VAT or LEI. We return a single verdict your code can branch on, backed by four independent signals. Each is checked against official, free-to-use registers and the live web; one being unreachable lowers confidence, it never silently passes.
It’s a registered company
The claimed entity resolves in an official register — VIES (EU VAT), GLEIF (LEI), Companies House (UK), INSEE SIRENE (FR) and more. A name with nothing behind it doesn’t pass.
Its address resolves to a real place
The registered business address resolves against the Acuris engine across 240+ countries — to point-coordinate level where the data allows. A fabricated or non-existent address is a tell.
That company controls this domain
The signal a KYB call alone can’t give you: does the storefront’s domain actually belong to the entity it claims to be? A perfect clone can copy a real company’s VAT and address — it can’t own the brand’s domain. This is where impersonators get stuck: they can copy the data, but they can’t reach a clean proceed.
Not on a sanctions list
The entity (and, where known, its owners) are screened against the EU, OFAC, UN, UK OFSI, Swiss SECO, French Trésor and Australian DFAT lists. A hit is a name match for human adjudication — never an automatic determination.
One decision, machine-readable
The four signals collapse into a recommended_action (proceed / review / do_not_proceed / block), a buyer_warning and a 0–100 score. Wire it straight into your agent’s pre-pay step.
Unreachable never means safe
If a register is down, the verdict degrades to review; if the storefront itself can’t be reached, it returns do_not_proceed — an unverifiable storefront is never treated as safe. The drop-in guardrail blocks payment on anything that isn’t a clean pass.
What it catches — and what it doesn’t
Honesty first, because a trust signal you can’t trust is worse than none. This verifies a storefront is a real, registered, sanctions-clean legal entity that controls the domain it’s selling from. It is not a safe-to-buy guarantee, a delivery guarantee, or a certification.
It catches
- Cloned & impersonating storefronts — a lookalike domain that isn’t owned by the brand it copies (the live #1 threat to shopping agents).
- Fabricated stores — a slick site with no real company, VAT or address behind it.
- Dead-brand revivals — a defunct or hijacked name brought back on a fresh domain to collect cards.
- Parked / newly-registered lookalikes — domains spun up days before a campaign.
- Sanctioned sellers — surfaced for adjudication against seven official lists.
It does not
- Guarantee a real, verified merchant will actually ship — non-delivery by a genuine company is a civil dispute, not impersonation.
- Defend against prompt injection of your agent — that’s agent integrity, a different layer.
- Detect buyer-side fraud or stolen-card use — that’s the payment network’s lane.
- Issue a certification, seal or compliance artefact — it’s a risk signal, not an accreditation.
- Fabricate a pass when it can’t bind: where a domain’s ownership can’t be cryptographically tied to the entity (e.g. GDPR-redacted WHOIS), the verdict degrades to
review— weaker proxy signals (domain age, registrar reputation) only ever raise flags, they never grant a clean bind.
The honest edge case: a sophisticated clone can display a real company’s genuine VAT and address on a fresh lookalike domain — and the register checks will pass it — but without the domain binding it lands at review, never a clean proceed. That is exactly why the website_bound signal exists: the binding between the entity and the domain is the one thing the impostor can’t fake. Read the four signals together, not in isolation.
Who puts this in the loop
AI shopping-agent platforms
You carry the reputational and liability exposure when your agent pays a scammer. One call before checkout turns “the agent bought from a clone” into “the agent flagged the clone and stopped.”
PSPs & agentic payment networks
Zero-liability covers unauthorized use — an agent-authorized payment to a fake merchant isn’t that. A merchant-legitimacy signal at the moment of pay is the slot the agentic-commerce protocols leave open.
Marketplaces & onboarding (DSA, KYBC)
Run it first-pass at seller onboarding — confirm the business is real and owns its storefront before it lists. Route weak scores to manual review instead of letting a phantom go live.
Brand & anti-counterfeit teams
Feed it suspect domains to confirm, at register level, whether a storefront trading on your name is actually you — evidence you can act on, not a hunch.
Drop it in before checkout
One HTTPS call, clean JSON, or the open-source guardrail that wraps your agent’s pay step and fails closed. (This is a real, documented case, not a hypothetical — see below.)
# One call: is this storefront who it claims to be? curl -X POST https://api.acuris-geo.com/storefront-verify \ -H "X-Acuris-Key: $ACURIS_API_KEY" \ -H "Content-Type: application/json" \ -d '{"domain":"russellbromleylondon.com", "name":"Russell & Bromley Limited","country":"gb"}' # → { # "domain_bound": false, // the clone tell # "entity_resolves": false, // owner in administration # "buyer_warning": "Likely a clone of a defunct brand — not the official store", # "recommended_action": "review", // not a clean pass → the guard won't pay # "score": 25 # }
Prefer a guardrail? pip install acuris-agent-guard gives you a one-line pre-pay wrapper for LangChain, CrewAI, Anthropic computer-use or any MCP agent — it blocks payment on anything that isn’t a clean pass. See it on GitHub.
This isn’t hypothetical — it already happened
In June 2026, national press reported that ChatGPT had recommended cloned Russell & Bromley shops to UK shoppers — a real luxury brand, a plausible storefront, deep “closing sale” discounts. Content-based checks saw nothing wrong, because the brand itself checks out: Russell & Bromley Limited (Companies House 00512958) is real, but it’s in administration, and its own domain now redirects to Next. No live store is verified as belonging to it — which makes every “Russell & Bromley” storefront on the internet clone-bait by construction, no matter how good it looks. A register + trademark lookup catches what content-based detection can’t.
See the full breakdown with dated receipts — the Companies House filing, the UKIPO trademark record, and the live HTTP redirect — on the Russell & Bromley case study →
Priced individually, on request
Agent Storefront Guard isn’t sold self-serve or bundled into a shared credit pool — tell us your call volume and we’ll quote a price.