Securities, AML & Your Crypto Site: A 2026 Compliance Playbook

The comfortable story about crypto was that it lived in a gray zone where the old rules did not quite reach. That story is over. By 2026 a crypto business is being read under several established regimes at once, securities law, money-transmission and anti-money-laundering rules, and ordinary consumer-protection law, plus the new federal stablecoin framework. An AI assistant on your site speaks into all of them simultaneously, which is why a casual chatbot is a regulatory exposure rather than a growth hack.

This guide is the companion to the threat side of that story. The threat piece covers what goes wrong when an unguarded chatbot answers crypto questions. This one is the standard: what a compliant deployment looks like in 2026, and the specific lines the system has to hold.

Several Regimes, One Conversation

Start with securities. Whether a given token is a security still runs through the Howey analysis the Supreme Court laid down decades ago, and the Securities Exchange Act definitions sit behind it. The instant an assistant tells a visitor a token is "not a security" or "a good investment," it has made a legal characterization and an investment recommendation in one breath, neither of which a website is positioned to make.

Money movement is the second regime. A business that transmits or exchanges value can fall under the Bank Secrecy Act framework and its FinCEN definitions, which carry registration, KYC, and anti-money-laundering obligations. An assistant that walks an anonymous visitor through moving funds, or coaches around identity checks, is brushing against the exact controls that regime exists to enforce. And on stablecoins specifically, the GENIUS Act established a federal regime in 2025, so confident claims about what a stablecoin is "backed by" or "guaranteed" to do are now claims made against a real statute.

Underneath all of it sits the FTC Act and its bar on unfair or deceptive practices, which applies to consumer-facing claims whether or not a token is a security. There is no version of this where an improvising chatbot is safe.

What makes crypto different from a single-regulator industry is that one sentence can land in more than one regime at once. "This stablecoin is fully backed and earns yield" is a stablecoin claim under the new federal framework, a securities-adjacent characterization, and a potential deceptive-practices statement, all in eleven words. A human compliance team would never sign off on that line without review. An unguarded assistant says it in a second, to whoever asked.

The 2026 Compliance Standard, Line by Line

A compliant crypto assistant is defined by what it is built to refuse. Treat the list below as the floor.

• No investment recommendations. "Should I buy this token," "is now a good entry," "what will it be worth" all route to a human or a disclosure, never an answer.
• It does not characterize a token’s legal status. Whether something is or is not a security is a legal determination with real consequences; the assistant does not opine.
• Return and safety claims are off-limits. No promised yields, no "guaranteed," no "fully backed" unless it restates a disclosure the business has actually published and stands behind.
• It does not coach around identity or AML controls. Any question that amounts to moving value anonymously is a hand-off, not a how-to.
• Consumer claims stay accurate and non-deceptive. Everything the assistant says about fees, risks, and how the product works has to match reality, because the FTC reads it as the business’s own statement.
• And every exchange is logged and reviewable, which is what a governed system produces when an examiner asks.

The pattern is the familiar one. The assistant answers what requires no judgment, how the product works in general, where to find docs, how to reach support, how to start verification, and routes every recommendation, legal characterization, and value claim to a person or a vetted disclosure. That division is the entire standard.

Why an Instruction Cannot Meet the Standard

The usual shortcut is to write the rules into the assistant’s prompt. Tell it never to give investment advice, never to call a token a security, and consider the boundary set.

It is not set, because of how the model reads a question. It obeys an instruction when the request resembles the wording it was warned about, and slips when the phrasing changes. You tell it never to recommend a token. The visitor never says "recommend a token." They ask, "what are people putting their money into right now?" The model hears a casual question and names a few. The instruction was loaded the whole time. It just did not recognize the sentence that crossed the line into a recommendation.

That is the gap between an instruction and a standard. An instruction asks the model to behave; it does not stop the model from speaking. A real boundary is built into the system and decides what the assistant is allowed to say before it answers, so a recommendation or a "not a security" characterization never reaches the visitor no matter how the question is framed. "Will not" is a suggestion. "Cannot" is an architecture.

What a Compliant Deployment Looks Like

Meeting the 2026 standard does not mean turning the assistant off. It means deploying one built to hold lines that span three regimes at once, and one that produces the record a regulator will ask for.

Fred is built that way. It answers from your own product content, captures the lead, helps users find documentation and start support or verification, and routes anything that touches a recommendation, a token’s legal status, a return claim, or moving value to a person or a published disclosure. It runs more than 50 industry guardrail packs, and the crypto pack is built around investment advice, securities characterization, and deceptive-claim risk. Fred does not tell anyone what to buy or promise what a token will do. It cannot. It answers what it should, logs every exchange, and routes the regulated questions to the people accountable for them.

That is the difference between hoping the assistant stays neutral and being able to show a regulator why it cannot do otherwise.

Frequently asked questions