Smart Contract Liability in 2026: Who Is Responsible When the Code Fails?

This is a follow-up to our earlier analysis of smart contract enforceability. That piece asked whether code can be a contract. This one asks the harder question — the one that actually shows up in a complaint: when the code executes exactly as written and someone loses $200 million anyway, whose name is on the caption?

The Premise Almost Everyone Gets Wrong

Liability does not attach to bytecode. An EVM contract is not a person, has no assets a court can reach beyond what sits in its own storage, and cannot be served. It has never been true that plaintiffs and prosecutors were stuck without a defendant. The real question — the one that determines outcomes — is which theory of attribution the government or a plaintiff uses to walk from the deployed contract back to a human being or an entity.

There are four such theories in active use in the United States right now: control-based regulatory liability, securities intermediary liability, unlicensed money transmission, and — the one builders systematically underprice — general partnership liability among token holders. Each has a different trigger, and critically, each attaches to a different layer of the stack.

The Five-Layer Liability Stack

Any serious risk analysis has to disaggregate the protocol into layers, because the law does:

1. Deployers and core developers — the parties who wrote and pushed the contracts. Exposure runs through fraud, money transmission, and aiding-and-abetting theories.
2. Governance participants — token holders who vote. Exposure runs through unincorporated-association and general-partnership theories.
3. Interface operators — the front-end, the router, the order-flow business. This is where broker-dealer, exchange, and futures commission merchant theories land hardest.
4. Infrastructure and service providers — oracles, relayers, sequencers, RPC endpoints, auditors. Mostly untested; negligent misrepresentation and aiding-abetting are the live theories.
5. The capital layer — venture investors, delegates, and multisig signers. This layer thought it was passive. Recent civil litigation says otherwise.

Mapping Failure Modes to Legal Theories

Technical failure	Primary US legal theory	Layer exposed
Reentrancy / arithmetic / logic bug	Negligence; breach of implied warranty; securities fraud if risk was misrepresented	Developers, auditors
Oracle price manipulation	CEA §6(c)(1) manipulation; wire fraud (against the attacker)	Attacker; possibly oracle provider
Flash-loan governance capture	Breach of fiduciary duty; partnership liability among voting holders	DAO, delegates
Admin key compromise or misuse	Conversion; §20(a) control person liability; fraud	Key holders, multisig signers
Upgrade that alters economics	Howey “efforts of others”; breach of contract	Developers, governance
Sanctioned or illicit usage	IEEPA; 18 U.S.C. §1960; BSA	Developers, front-end

Federal Enforcement: Four Agencies, Four Theories

CFTC: Control Is the Trigger

The CFTC has been the most doctrinally aggressive agency here, and its theory is the cleanest: if you control the thing, you are the operator of the thing. In CFTC v. Ooki DAO, No. 3:22-cv-05416 (N.D. Cal.), the Commission charged the DAO itself as an unincorporated association, arguing that token holders who voted on governance proposals were members of that association and therefore jointly liable. Judge Orrick rejected the DAO’s challenge to that structure, and the CFTC obtained a default judgment on June 8, 2023, including a roughly $643,542 penalty and an order shutting down the front-end.

Ooki is the single most important precedent in this area, and builders consistently misread it. The holding is not “DAOs are illegal.” The holding is that voting is conduct, and conduct creates membership, and membership creates liability. The predecessor entity (bZeroX) had explicitly told users it was transferring control to a DAO to become “enforcement-proof.” The CFTC quoted that back at them.

The Commission followed with a set of simultaneous settled actions on September 7, 2023, against Opyn, ZeroEx, and Deridex — all DeFi protocols, all charged with offering leveraged retail commodity transactions and failing to register. Notably, the CFTC targeted the interfaces and the operating companies, not the immutable contracts. That distinction is the whole ballgame.

SEC: Howey Plus Intermediary Status

Applying SEC v. W.J. Howey Co., 328 U.S. 293 (1946), to a protocol requires care. The four prongs — investment of money, common enterprise, expectation of profits, derived from the efforts of others — rarely fail on the first three. The fight is always prong four, and prong four is a question about code architecture, not marketing.

Consider what an upgrade authority actually looks like:

// UUPS proxy pattern — legally, this is the Howey prong-four exhibit
function _authorizeUpgrade(address newImplementation)
    internal
    override
    onlyOwner
{}

// Whoever satisfies onlyOwner can rewrite fee logic, reward
// curves, and withdrawal rules for every existing holder.

That modifier is not a technical detail. It is documentary evidence that profits depend on the managerial efforts of an identifiable party. A protocol with a live upgrade key, a treasury-controlled fee switch, and a core team publishing roadmaps has assembled the elements of prong four in its own repository. Conversely, genuinely immutable contracts with no privileged roles make the “efforts of others” argument substantially harder for the Commission to sustain — which is the actual legal content of “sufficient decentralization,” a phrase that has never appeared in a holding but describes a real evidentiary posture.

Two SEC actions matter most for builders. In the settled BarnBridge DAO matter (December 2023), the Commission charged a DAO and its founders over structured yield products, extracting disgorgement and effectively demonstrating that a DAO wrapper does not absorb founder liability. And in SEC v. Rari Capital (September 2024), the Commission’s order emphasized that the protocol was marketed as autonomous while in fact requiring substantial ongoing managerial intervention. Rari is the cleanest statement of the rule builders need: the gap between your decentralization claims and your actual admin surface is itself the violation.

DOJ and FinCEN: The Money Transmission Exposure

This is where developer criminal risk concentrated. 18 U.S.C. §1960 criminalizes operating an unlicensed money transmitting business, and prosecutors have argued that writing and maintaining software that moves value can qualify. That theory sits in tension with FinCEN’s own guidance — FIN-2019-G001 (May 9, 2019) — which states that a person who merely supplies anonymizing software, as distinct from operating an anonymizing service with control over funds, is not a money transmitter.

The Tornado Cash prosecutions in the S.D.N.Y. tested exactly that line, as did the Samourai Wallet case. The policy environment shifted materially in April 2025, when the Department of Justice issued a memorandum narrowing digital-asset charging practice and disbanding its dedicated crypto enforcement unit. Builders should treat that as a prosecutorial policy change, not a change in law — §1960 remains on the books, memoranda are revocable, and state prosecutors are unaffected. The proposed Blockchain Regulatory Certainty Act, which would create a statutory safe harbor for non-custodial developers, remains the durable fix and remains unenacted as of this writing.

The operative test is custody and control. If your contracts never take possession of user funds in a way you can unilaterally direct, you have the strongest available defense. If you hold an emergency withdrawal key, you have undermined it.

OFAC: The Immutability Holding

Van Loon v. Department of the Treasury, 122 F.4th 549 (5th Cir. 2024), is the most builder-favorable federal decision in this entire area. The Fifth Circuit held that immutable smart contracts are not “property” within the meaning of IEEPA, because no one can own or control them — and OFAC therefore lacked authority to designate them. Treasury removed the Tornado Cash contract addresses from the SDN List in March 2025.

Read carefully, Van Loon is a narrow but powerful proposition: immutability is a jurisdictional fact. It does not immunize developers, front-ends, or relayers, none of whom are immutable. But it establishes that the government’s sanctions authority runs out at the point where human control does.

The Sleeper Risk: General Partnership Liability

Regulatory enforcement gets the headlines. Private civil litigation is the larger threat to individual net worth, because it carries unlimited personal liability.

In Sarcuni v. bZx DAO, No. 3:22-cv-00618 (S.D. Cal. Mar. 27, 2023), the court held that plaintiffs plausibly alleged the DAO was a general partnership under California law, allowing negligence claims against members to survive dismissal. The underlying facts were a phishing compromise of a developer that drained roughly $55 million — a security failure, converted into a partnership-duty claim.

In Samuels v. Lido DAO, No. 3:23-cv-06492 (N.D. Cal. Nov. 18, 2024), Judge Chhabria went further, holding that the DAO could plausibly be a general partnership and that institutional investors who participated in governance could be liable as partners. That ruling detonated a comfortable assumption across the venture community: that buying and voting governance tokens is passive investment. Under partnership law, participation in the management of a business for profit is the definition of a partner, and partners are jointly and severally liable for partnership obligations.

Note what this means mechanically. A DAO with no legal wrapper is not a legal vacuum — it is a default general partnership, the worst available entity form, in which every voting member is exposed to the full liability of the enterprise.

Protocol Versus Front-End: Where the Line Actually Falls

The most reliable structural defense in DeFi is the separation between immutable protocol contracts and the operated interface. The case law supports it. In SEC v. Coinbase, Judge Failla’s March 2024 opinion dismissed the Commission’s claim as to Coinbase Wallet, reasoning that providing self-custodial software that routes transactions is not, without more, brokerage. And the SEC’s investigation of Uniswap Labs — opened with a Wells notice in April 2024 — was closed without enforcement in early 2025.

But the separation only works if it is real. The CFTC’s Opyn and ZeroEx orders show the agency going straight at interfaces. Front-ends that geoblock, screen addresses, filter tokens, or route order flow for fees are performing intermediary functions, and intermediary functions attract intermediary regulation. The uncomfortable irony is well known and unresolved: the compliance measures that reduce AML risk simultaneously build the record of control that establishes broker or FCM status.

State and International Overlay

Entity law. Wyoming’s DAO LLC statute (W.S. §17-31-101 et seq.) and its Decentralized Unincorporated Nonprofit Association Act (effective July 1, 2024) both provide members with limited liability — the direct antidote to bZx and Lido. A Cayman foundation company or a Marshall Islands non-profit LLC serves similar ends for offshore structures. Any protocol with a treasury and active governance that has not adopted a wrapper is accepting general partnership exposure as a design choice.

Money transmission. State licensing runs independently of federal policy. New York’s BitLicense (23 NYCRR Part 200) and roughly 49 state MTL regimes apply on their own terms, and a state DFS or AG action is unaffected by any DOJ charging memorandum.

Enforceability. E-SIGN (15 U.S.C. §7001) and state UETA enactments already make electronic and automated contracting valid; Arizona, Tennessee, and Wyoming have gone further with express smart contract recognition. The persistent problem is not validity but gap-filling — a smart contract executing on a manipulated oracle price has performed correctly and produced a result no party bargained for.

International, with the US angle. MiCA (Regulation (EU) 2023/1114) became fully applicable on December 30, 2024 but largely carves out fully decentralized arrangements, deferring the DeFi question to a subsequent review. US firms serving EU users through an operated front-end should assume CASP authorization analysis applies to the interface even where the protocol is out of scope. The UK Law Commission’s 2021 conclusion — that existing contract law accommodates smart contracts without new legislation — remains the leading common-law position and is broadly consistent with how US courts have approached the question.

Risk Assessment

Actor	Exposure	Severity	Mitigation strength
Core developer with live admin keys	§1960, securities fraud, control person	High	Weak — control is the trigger
Developer of immutable, non-custodial contracts	Aiding and abetting; residual §1960	Moderate	Strong — Van Loon, FinCEN 2019 guidance
Front-end operator collecting fees	Broker/FCM/exchange registration, BSA	High	Moderate — structural separation
DAO with no legal wrapper	General partnership; unincorporated association	Severe	Weak — bZx, Lido, Ooki
DAO in Wyoming DAO LLC or DUNA	Entity-level only	Moderate	Strong — statutory limited liability
Governance voter / delegate	Partnership liability; fiduciary duty	High and underpriced	Weak absent a wrapper
Auditor	Negligent misrepresentation to third parties	Low but rising	Untested — disclaimers largely unlitigated

Key Takeaways for Builders

1. Your admin surface is your liability surface. Enumerate every privileged role, upgrade path, and pausable function. That list is the prosecution’s exhibit and the plaintiff’s roadmap. Reduce it deliberately, and document the reduction with timestamps.
2. Never claim more decentralization than your code delivers. Rari establishes that the gap between the claim and the architecture is independently actionable. Marketing copy is discoverable.
3. Wrap the DAO before it holds a treasury. A Wyoming DAO LLC or DUNA costs a few thousand dollars. Unlimited joint and several liability for every voting member does not have a price.
4. Separate the interface from the protocol, in fact and not just on a slide. Distinct entity, distinct counsel, distinct revenue, and an honest assessment of whether fee-taking and order routing have made the interface an intermediary.
5. Custody is the criminal-law fault line. If your contracts can never unilaterally direct user funds, your §1960 defense is materially stronger. An emergency withdrawal key trades a security benefit for a criminal exposure — make that trade consciously.
6. Treat prosecutorial policy as weather, not climate. The 2025 DOJ posture is favorable and revocable. Statutes of limitation outlast administrations.
7. Adopt a published whitehat safe harbor and a documented incident plan. Ad hoc negotiation with an exploiter during a live drain is how good-faith recovery efforts become obstruction allegations.

The Genuinely Unsettled Questions

Three questions have no reliable answer, and counsel who claims otherwise is guessing. First, whether “code is law” survives as a defense to fraud when a trader exploits a protocol precisely as written — post-trial rulings in the Mango Markets litigation have cut against the government’s fraud theories, and appellate resolution is pending. Second, whether validators and block builders owe duties to transaction senders, which the MEV prosecutions have raised without settling. Third, whether an auditor’s disclaimer survives a Restatement §552 negligent-misrepresentation claim by a third-party depositor who relied on the report. That case is coming.

The through-line across everything above is simple enough to state in one sentence: US law locates liability wherever it finds control, and every line of privileged code you ship is an admission of where that control sits.

---

This content is for informational purposes only and does not constitute legal advice. Consult with a qualified US attorney for guidance specific to your situation.