The pitch arrived in mid-October, five weeks before Black Friday, wrapped in four numbers: orders up fourfold, automated fraud doubled, headless checkout ten times faster than screen scraping, token efficiency five times better. Every one of those figures is Forter's own, verified by nobody but Forter. The company said so itself. That admission is the most honest thing in the release, and it is also the thread worth pulling — because what Forter is actually selling has almost nothing to do with those numbers.
Forter is a merchant-side risk vendor. It does not hold a license, does not settle funds, does not touch a bank core. Its new product, the second phase of an "Agentic Commerce Engine," sits between an AI agent trying to buy something and the merchant deciding whether to say yes. Single integration, no code, and — this is the part that matters — the ability to set the conditions of each transaction at the moment it happens. On paper that is decision support. In practice it inches toward payment facilitation, and the distinction between the two is the entire regulatory question nobody in the announcement wanted to name.
I have audited enough of these architectures over the years to recognize the shape. The orchestration layer that approves or declines a checkout without holding a payment license must, by legal structure, attach itself to a licensed PSP or merchant of record. The moment Forter starts making the final call on admission rather than offering a recommendation, regulators can reclassify it as a payment participant. That is why the language stays deliberately soft. The company wants to be the advisor, never the counterparty. Whether that framing survives its first serious incident is a different matter.
The protocol layer is where the real hierarchy lives, and Forter is not in it. The announcement maps a tidy three-tier world: Cequence at identity, Forter at orchestration, Harbor at insurance. Four protocols float above — Visa's TAP, Google's AP2, the OpenAI-Stripe ACP, and Skyfire. Forter claims to be protocol-agnostic, which sounds like strength and is actually an admission. It cannot become the protocol, so it adapts to four of them, all still evolving, none converged. The cost of that position is not research and development. It is a permanent adaptation tax paid to standards it does not control.
Notice what that map leaves out. The card networks, Google, Stripe, and OpenAI set the rules. Acquirers move the money. None of them appear in Forter's diagram, because accepting the diagram means accepting Forter as infrastructure rather than as a small company selling plugins to infrastructure. It is good public relations cartography, and it should be read as exactly that.
The disclosed numbers deserve the same skepticism. Fourfold order growth sounds like market validation until you place it beside the other figure in the same release: fewer than eleven percent of small merchants say they are prepared. Fourfold from a very small base is still a very small number. The fraud figure doubling in two months is the more interesting data point, and it cuts both ways. If automated fraud is rising faster than legitimate agent volume, the economics of this channel at the merchant level may currently be negative. That would put a hard ceiling on 2027 expansion regardless of how elegant the orchestration layer is.
Then there is the structural fracture in the technical story. Forter distinguishes legitimate AI assistants from fraudulent abuse bots using behavioral signals and the transaction graph from its own network, which has genuinely grown alongside agent volume. Behavioral detection cannot tell a legitimate agent from a legitimate agent that has been hijacked, because the two behave identically. The only real answer to that is cryptographic attestation of agent identity, and that layer belongs to Cequence, Visa, and Skyfire. Forter's risk verdict is, at the top of the chain, a forgeable input. It sees what flows through its merchants and makes probabilistic calls on it, which is useful work. It cannot prove who an agent is, which is the work that will define the category.
One indicator in the release should trouble anyone reading it as a growth story. The wording distinguishes between functions live for existing customers and functions open to interested prospects. The first is a free extension of existing contracts — pure cost, zero incremental revenue, defensive spending to stop churn. The second is a new line. In the short run, agentic functionality is insurance against losing accounts, not an engine of income.
And here the deeper exposure surfaces. Forter built its reputation on chargeback guarantees — it underwrote fraud losses as a differentiator. If that guarantee extends to agentic orders, then "automated fraud doubled in two months" is not an industry observation. It is a signal pointed directly at Forter's own profit and loss. Which explains the urgency of routing liability outward, toward an insurance layer and toward the AI providers that ninety-three percent of merchants say should carry the risk. Forter is trying to convert itself from underwriter back into software vendor, and it is doing so five weeks before peak season.
That timing is not about holiday traffic, whatever the release says. Harbor came out of stealth the same day with seed funding. This sector is raising money right now. A product launch is a valuation narrative, and Black Friday is the prop.
Compliance, meanwhile, has not caught up. Agentic commerce currently runs on private ordering: liability is set by Visa's and Stripe's and Google's terms, plus insurance contracts, rather than by regulators. That arrangement holds until the first mass incident involving hijacked agents and stolen cards, after which rulemaking arrives as catch-up and the early architectural choices may become instantly non-compliant. Europe is the hardest market here, and the one the release avoids. Strong customer authentication under PSD2 presumes the cardholder authenticates personally. Proxy shopping, by definition, conflicts with that. Without an exemption path for delegated authorization, agentic commerce in Europe stays structurally capped. Skyfire quietly drags the whole thing into stablecoin settlement, which means MiCA, US stablecoin law, and the Travel Rule enter the picture — a compliance branch the announcement never mentions and cannot avoid.
Europe is also where the small-merchant problem turns into policy. Small merchants are the most exposed party in this architecture: most likely to receive agent traffic passively, least able to absorb a bad decision, least able to negotiate terms. They are precisely the constituency a regulator would protect, and protecting them means pushing liability further up toward AI providers and large platforms. That direction helps Forter shed risk. It also invites mandatory technical standards and compliance costs that no orchestration vendor can dodge.
I keep coming back to the one asset Forter has that is genuinely hard to copy: its transaction graph, fed by real order flow. That is a real moat, of medium strength. The integration moat is weaker than it looks, because no-code integration lowers barriers for everyone, including the next entrant. There is no license moat. And the network effect is one-sided. True two-sided effects require agent platforms to treat Forter as a default route, but platforms integrate identity registries, not risk vendors.
Where does this settle? A middleware company with real data, a defensive product, a guarantee book that may be quietly bleeding, and a window that stays open only as long as four protocols remain unreconciled. Twelve to twenty-four months, most likely. When one standard wins, merchants stop needing a protocol-agnostic translator, and Stripe — which co-writes ACP and sells Radar — is simultaneously Forter's upstream rulemaker and downstream competitor. That dual role is the structural threat, and no competitor named Riskified or Signifyd comes close to it.
The question worth carrying into the holiday quarter is not whether the engine holds up under load. It is who signs the loss when an agent that passed every check turns out to have been someone else's puppet — and whether the answer is a software vendor's problem, an insurer's problem, or the beginning of a rule that makes all of this someone's regulated business.
