Tornado Cash's Real Trial Is About Who Owns the Code: Inside the DOJ–Treasury Split

On the morning of October 6, 2026, FinCEN published a withdrawal notice that erased a reporting obligation the crypto industry had spent two years fighting. Its language was careful but unmistakable: the agency acknowledged a category of "legitimate financial privacy." Less than a day later, in the Southern District of New York, federal prosecutors stood before Judge Katherine Failla and argued that Roman Storm should face a jury for building the very tool the Treasury had just declined to regulate as inherently illicit. Two branches of one government, one week, opposite postures. The regulatory thermometer fell; the prosecutorial barometer climbed. I have audited enough failed projects to distrust clean narratives, and this one is anything but clean. But if you hold privacy assets, or you build them, the contradiction is not noise. It is the signal — and it points to a question no compliance memo has answered: when a developer ships immutable code, who exactly is the money transmitter? Tornado Cash is a zero-knowledge mixer on Ethereum. Users deposit funds, and zk-SNARK proofs let them withdraw the same amount to a fresh address without an on-chain link. For years it was the reference implementation for on-chain privacy — the first ZK mixer validated at scale. In 2022 the Treasury's OFAC sanctioned the protocol, and in 2023 the DOJ indicted co-founders Roman Storm and Roman Semenov. The charges are not about securities. They are about two statutes: 18 U.S.C. § 1960, operating an unlicensed money-transmitting business, and § 1956, conspiracy to launder money. The government's legal scaffolding leans on Bitcoin Fog, the centralized mixer whose operator, Roman Sterlingov, was convicted — a case now on appeal. Prosecutors have used that precedent to widen the Tornado Cash theory, arguing that running a mixer is itself a transmission business regardless of custody. Meanwhile, the DOJ's own internal guidance, including the Blanche memorandum, has signaled limits on certain charges — even as line prosecutors press forward. Storm has already been convicted on one count; the rest, including the venue challenge, remain live before Judge Failla. The decisive technical question is decentralization — and it is not philosophical. It is the hinge on which criminal liability swings. If Tornado Cash is genuinely automated peer-to-peer software, § 1960 may not reach it. If the founders retained operational control, the statute bites. The government's answer is the relayer. ZK mixers need a front end and relayers to submit withdrawal transactions, and someone operates those layers. Prosecutors point to that operational role as proof the team "built, maintained, and profited from" a business. In their framing, maintaining the pool is not a neutral act — it is extending credit to whoever withdraws. Under that logic, knowledge of illicit deposits, whether or not the developer benefited, is enough for conspiracy. The venue challenge narrows the question but sharpens it: did any transmission happen in Manhattan, or was a brief deposit by a single customer — Shakeeb Ahmed — enough to anchor jurisdiction? That is a breathtaking expansion, and Judge Failla said so. If maintaining any permissionless protocol makes its author a co-conspirator, then every DeFi team is one bad actor away from a federal indictment. Here is the detail most coverage missed: the TORN governance token may be the prosecution's quietest weapon. A governance token is evidence that a team retains control and upside — the exact elements that convert "software" into "business." The token that was supposed to decentralize the project becomes the exhibit that centralizes the blame. Code is law, but people are the context — and the context here is a governance token that proves someone was still steering. There is a second, quieter lesson in the split between the two agencies. FinCEN's retreat — and its explicit nod to legitimate privacy — suggests the administrative state is beginning to accept that privacy is a design requirement, not a crime. That is the opening privacy builders have waited for. But it only holds if the technology internalizes compliance: allowlists, sanctioned-address screening, proof that a user is not a sanctioned party, all baked into the circuit rather than bolted on at the front end. Community over coin, always — and the community that survives regulation will be the one that designs for it rather than pretending it does not exist. Based on my audit experience, I have learned to read whitepapers for red flags, not just bugs. The flag here is structural: a protocol can be immutable on-chain while its social and economic layers remain firmly under human control. The contract cannot be changed; the relayers, the front end, and the treasury can. Decentralization was never a binary — it was a spectrum, and the courtroom is now drawing the line. The market's instinct is to read FinCEN's withdrawal as a thaw. That is a mispricing. An administrative retreat removes a reporting duty; it does not grant immunity. FinCEN itself was explicit: administrative policy does not decide criminal cases. The two tracks run on separate rails. So the "policy easing" rally in privacy names is likely capped. Exchanges and institutional desks will not relist TORN or integrate mixers while a federal prosecution is live — the legal risk dwarfs the administrative relief. Expect liquidity to keep migrating offshore, where compliance-averse market makers are willing to touch the asset class, while domestic desks stay clear. Every unlicensed privacy protocol now carries a risk discount priced against the possibility that a jury decides developers can be liable for their users. Anonymity is a shield, not a lifestyle — yet a jury may struggle to tell a shield from a getaway car. The real verdict here will not be about one man or one mixer. It will define whether privacy is a right you can build or a feature you must license. Trust is the only protocol that matters — and right now, no one trusts the rules, because no one has written them. The question the next decade must answer is not whether privacy belongs on-chain. It is who gets to decide.