Talk to enough underwriters right now and a pattern shows up fast: AI is one of the hardest categories to place a merchant account for in 2026, and it's not because AI companies are unusually fraudulent. It's because the entire underwriting model banks use — look at what the business sells, how it's delivered, and what could go wrong — breaks down when the product is a model that generates its own output. Sponsor banks have spent years building risk frameworks around known categories. AI doesn't fit cleanly into any of them, and "we don't know how to categorize this" tends to resolve as a decline rather than a deeper look.

"How Do You Police It?"

That's the question underwriters keep coming back to, and it's a fair one. A traditional e-commerce business sells a fixed catalog — a bank can review the product list and know what it's underwriting. A generative AI platform doesn't have a fixed catalog. The output changes based on what the user types in, which means the bank is effectively underwriting an unbounded range of possible content it can't preview in advance.

This is where the nudity and age-verification problem shows up, and it's broader than it looks. A platform doesn't have to be adult-branded to end up producing content that lands it in adult-content territory — an image generator, a chatbot, or a video tool can be pushed by users into NSFW output regardless of what the platform intended, and banks have no reliable way to audit that risk before it happens. The same content-moderation exposure that makes adult platforms hard to bank (see our guide to AI dating and AI adult platforms and current age verification laws) applies, in a milder but real form, to almost any generative AI product a bank can't fully preview. A general-purpose AI tool doesn't get a pass just because it wasn't built for that use case — from an underwriting seat, unpredictable output is unpredictable output.

Free Trials Are the Riskiest Part of the Model

Most AI products launch with a free trial or a freemium tier, because that's how the category proves value before asking for a card. From an underwriting perspective, that's close to the worst possible billing structure. Trial-to-paid conversions generate a predictable spike in "I didn't mean to subscribe to this" disputes, and a brand-new AI startup running trials has no processing history yet to show how those disputes actually play out. The bank is being asked to underwrite the riskiest phase of the business model with the least amount of data to evaluate it by.

Funding Doesn't Solve the Problem — It Complicates It

AI startups raise unusually large rounds unusually early, often before there's meaningful revenue. That funding is frequently offered as evidence the business is legitimate and can absorb risk. Underwriters don't take it at face value for one specific reason: if the evidence of financial strength is liquid assets sitting in a different entity — a parent company, a holding company, an affiliated fund — the bank needs to see a clear, documented connection between that entity and the one applying for the merchant account. Money that isn't traceably tied to the applicant doesn't reduce the account's risk in the bank's eyes; it just raises a question about why it's being offered as if it does.

Large early funding rounds also mean many AI companies are burning through outside capital rather than running on their own processing revenue, which is a different risk profile than a bootstrapped business growing off actual sales — one that underwriters price and structure differently, and are often less familiar with.

The CBD Parallel

Banks have been here before. When CBD products became federally legal to sell after the 2018 Farm Bill, the regulatory reality was messier than the headline: the FDA hadn't cleared a path for CBD in food or supplements, card networks restricted the category regardless of state legality, and every bank was left guessing how much liability it was taking on. The result wasn't careful, case-by-case underwriting — it was blanket avoidance. Most banks simply declined the entire category rather than build the expertise to evaluate individual CBD merchants, and it took years and a wave of specialized high-risk processors before CBD had anything resembling normal access to payments.

AI is following the same shape for the same underlying reason: a fast-moving category outpacing the regulatory clarity banks want before they'll underwrite it, with enough individual businesses inside it that a handful of bad actors are enough to make the whole category feel radioactive. The difference is scale — there are far more AI companies chasing this gold rush than there ever were CBD brands, which means far more merchant account applications hitting the same wall at once.

What Actually Gets an AI Company Approved

None of this makes AI unbankable — it makes it a specialist category, the same way CBD eventually became one. The applications that get through underwriting tend to have a few things in common:

  • A documented content moderation approach. Output filtering, usage policies, and a real process for handling misuse — something concrete a bank can point to instead of taking "we monitor for abuse" on faith.
  • Traceable funding. If solvency is being demonstrated through funds held elsewhere, bring the documentation that connects that entity to the one applying — corporate structure, ownership, and the actual flow of funds.
  • A trial structure with limits. Usage caps, card-on-file verification before trial start, and clear renewal disclosures reduce the dispute spike that comes with unrestricted free access.
  • A processor that already underwrites AI. The same way CBD needed processors who'd built real expertise in that category, AI startups do better with high-risk processors who've already worked through these questions rather than a generalist encountering them for the first time.

Getting declined by a mainstream processor isn't a signal your business is unbankable — it's a signal you're early in a category still being figured out. Compare high-risk merchant account providers experienced with AI and generative platforms instead of taking the first decline as final.

Frequently Asked Questions

Why won't mainstream processors like Stripe or Square work with AI startups?

It varies by product, but generative AI platforms in particular carry output risk mainstream processors aren't built to evaluate — they can't preview what the product will generate the way they can review a fixed catalog of goods.

Does having a lot of VC funding help get approved?

Only if the funding is documented and traceable to the applying entity. Funding sitting in a separate but unconnected entity doesn't reduce underwriting risk on its own — it needs a clear, documented link to the business being underwritten.

Is every AI company treated as high-risk?

Not automatically, but most end up there by default because mainstream processors decline the category outright rather than evaluate individual businesses, similar to how CBD was treated in its early years.

What's the single biggest thing an AI startup can do to improve its odds?

Show a real content moderation and misuse-handling process. It's the single factor underwriters cite most often as the difference between "we can't police this" and an account they're willing to take on.