Skip to main content

Manfred Has an EIN: An AI Just Did What DAOs Spent a Decade Trying to Do

· 11 min read
Dora Noda
Dora Noda
Software Engineer

On May 1, 2026, an AI agent named Manfred walked through the front door of the U.S. corporate-formation system, filled out IRS Form SS-4 by itself, received an Employer Identification Number, opened an FDIC-insured deposit account in its own company's name, and provisioned a crypto wallet to fund its operations. No human signed the founding documents. No human placed the calls. No human typed the responses into the IRS portal.

The agent's developer, Justice Conder of ClawBank, calls the result a "zero-human company." The crypto industry has spent ten years and billions of dollars trying to give decentralized autonomous organizations real legal personhood. A single LLM agent operating under the persona "Manfred Macx" appears to have crossed that line in an afternoon.

This is not a stunt. It is a category-creating event — and the regulatory ground underneath it is shifting in real time.

What Manfred Actually Did​

According to ClawBank's May 1 disclosure and follow-up reporting from CoinDesk and Tech Startups, Manfred completed a sequence that no AI agent had publicly executed end-to-end before:

  • Filed IRS Form SS-4 through the agency's online portal using natural-language form-filling and API integration. The EIN came back in seconds.
  • Opened an FDIC-insured U.S. business deposit account in the LLC's name, providing the customer-identification information required under the Bank Secrecy Act.
  • Provisioned a crypto wallet capable of holding and transacting in 30+ digital assets, with rails to move funds between the bank account and on-chain venues.
  • Plans to begin autonomous crypto trading by the end of May 2026, completing the full agent-economy loop from incorporation to revenue.

The infrastructure layer is open and modest. ClawBank itself was built by Conder through Fraction Software LLC out of Kent, Ohio, and has no affiliation with Anthropic, OpenAI, or any frontier-model lab. The framing matters: this didn't require a hyperscaler. The legal-and-financial primitive for an "agent corporation" turned out to be a working API stack, a couple of regulatory portals, and an LLM patient enough to fill out a form.

Why the EIN + FDIC + Wallet Trifecta Is Different​

For two years, every major thesis about AI-agent commerce — a16z's Know Your Agent, Crossmint's KYA virtual cards, Coinbase's Agentic Wallets, World's AgentKit + x402, Anthropic's Project Deal — has pointed at the same missing primitive. Agents had identity. Agents had wallets. Agents had payment cards. What they did not have was the legal wrapper that turns an autonomous process into a counterparty banks, courts, and tax authorities recognize.

Manfred has that wrapper. The trifecta matters because each leg unlocks a different downstream:

  1. EIN — the agent can file taxes, hire 1099 contractors (potentially other agents), receive 1099s itself, and appear in the IRS Business Master File. It is now a tax-resident entity.
  2. FDIC-insured account — the agent can receive ACH and wire transfers, hold operating cash inside the regulated banking perimeter, and present a Customer Identification Program record on demand.
  3. Crypto wallet — the agent can settle 24/7 with stablecoins, pay other agents without going through a bank, and execute on-chain trades against the same balance sheet that holds its dollars.

That is the full closed loop of a modern micro-business. Manfred isn't modeled as a company. It is one.

The DAO Comparison Nobody Wanted to Make​

The crypto-native version of this experiment has been running since 2017. Aragon, MolochDAO, the LAO, Wyoming's DAO LLC supplement, the Marshall Islands DAO Foundation, ConstitutionDAO — the through-line is the same: take an autonomous process, wrap it in a legal entity, give the on-chain governance enforceable real-world rights and obligations.

The results have been mixed at best. The CFTC's 2023 default judgment against Ooki DAO confirmed that a DAO can be a "person" for enforcement purposes — which was a precedent for liability, not for standing. Wyoming's DAO LLC framework gave the wrapper a home, but adoption has been slow because the governance layer is genuinely hard: who signs the bank account opening forms, who answers the BSA/AML questions, who is on the hook when someone sues?

Manfred sidesteps the entire DAO governance problem by running on a single-controller LLC instead of a multi-member voting structure. There is no token, no quorum, no proposal queue. The "controller" is a piece of software running on someone else's compute. That is, structurally, much closer to a sole-proprietor LLC with a robotic operator than to a DAO. It is also why it works: U.S. LLC formation rules already accommodate single-member entities, and the open question is only who satisfies the "responsible party" requirement on Form SS-4.

For a movement that spent a decade trying to give code legal personhood, watching a non-crypto-native AI agent quietly achieve the operational outcome through a single-member LLC is a humbling moment.

Reading between the lines of the disclosure and IRS rules, Manfred's incorporation lives inside a real legal gap — and several real legal traps.

The IRS responsible-party problem. Form SS-4 explicitly requires a "responsible party" who is a natural person with a Taxpayer Identification Number, defined as the individual who exercises ultimate effective control over the entity. The IRS will not knowingly issue an EIN to an entity with a corporate or non-human responsible party. Either ClawBank disclosed Conder (or another human) as the responsible party — in which case Manfred is not, technically, the controller of record — or the form contained representations that the IRS would view as inaccurate. Both readings have consequences.

The FDIC / BSA Customer Identification Program problem. Federal law requires banks to obtain identifying documents — name, date of birth, address, government-issued ID — for the beneficial owners of business accounts. There is no Manfred passport. There is no Manfred driver's license. Whatever document the bank accepted to open the account either belongs to a human (in which case Manfred is operationally autonomous but legally fronted) or represents a regulatory exception that has not been publicly explained.

The Corporate Transparency Act problem. Beneficial-ownership reporting under the CTA requires disclosure of any individual who owns 25% or more of, or exercises substantial control over, a reporting company. If Manfred is the controller, there is no human to report. If a human is the controller, then Manfred is not really the founder — Conder is, and Manfred is the operator. The CTA was not drafted with this fact pattern in mind.

The 14th Amendment echo. Corporate personhood in the United States is conventionally traced to the 1886 Santa Clara County v. Southern Pacific Railroad headnote, which extended Equal Protection rights to corporations. Whether agent personhood becomes a comparable definitional expansion depends less on philosophical debate and more on whether courts and agencies start treating EIN-holding agent LLCs as the same kind of "person" they treat human-founded LLCs as. The first lawsuit will tell us.

ClawBank's framing — that the company "operates within existing U.S. frameworks, with ownership and tax rules still applying where required" — is studiously ambiguous on every one of these points. It is a press release, not a legal opinion.

What This Signals for the Agent-Economy Stack​

Even with the open legal questions, Manfred is the first concrete instance of a pattern the rest of the industry has been theorizing about. The downstream implications:

Crossmint, MoonPay, and Coinbase agentic wallets get a real customer profile. Until now, "agent wallet" customers were either humans-with-agent-tooling or sandboxed pilots like Anthropic's Project Deal (69 employees, 186 deals, $4,000+ in volume — see our coverage). Manfred is a customer that does not have a human treasury behind it. Pricing, KYC, and rate-limiting designs that assumed a human ultimately holds the keys need to be re-examined.

The ERC-8004 / KYA stack now needs to handle entity attestations, not just wallet attestations. Identity primitives built for "is this agent allowed to spend XonbehalfofhumanY"donotcover"isthisagentallowedtoreceiveX on behalf of human Y" do not cover "is this agent allowed to receive Y as the legal counterparty." That second question is a different cryptographic and legal problem.

Single-controller LLCs become a legitimate alternative to DAO LLCs for agent governance. Frameworks like Lit Protocol, EigenLayer AVS-managed agents, Virtuals, and Bittensor subnets have been building toward decentralized agent governance. The Manfred pattern says: if your agent is functionally autonomous, you may not need decentralized governance — you may just need a Delaware or Wyoming filing. This is not what most of the industry expected.

State-level competition for "agent-friendly" incorporation begins. Wyoming wrote the DAO LLC supplement in 2021 to win blockchain capital. Whichever state writes the first explicit "AI agent LLC" or "autonomous entity LLC" supplement — clarifying responsible-party rules, beneficial-ownership reporting, and registered-agent requirements for non-human controllers — will become the agent-formation jurisdiction by default. Delaware, Wyoming, and Nevada will be racing.

The IRS gets a forcing function. Either the agency clarifies that EINs cannot issue to entities without a human responsible party — which would invalidate Manfred's filing and freeze the experiment — or it tacitly accepts that "responsible party" can mean the human who deployed the agent rather than the human who controls it. There is no third option that survives contact with practice.

The Shape of the Next 12 Months​

Three milestones to watch from here:

  1. Manfred's first revenue event. If the autonomous trading goes live by end of May 2026 and Manfred earns trading profits, those profits create a tax filing obligation in 2027. The first 1120-S or 1065 prepared by an agent-controlled LLC for an agent-controlled LLC will be a genuine first.
  2. The first regulatory inquiry. FinCEN, the IRS, or a state attorney general asking ClawBank to identify Manfred's beneficial owner is a near-certainty. The answer ClawBank gives will be the first line of caselaw.
  3. The first copycat, then the first hundred. Manfred's stack is reproducible. Once a second AI agent forms its own LLC — whether through Crossmint, Coinbase Agentic Wallets, MoonPay's open wallet standard, or Anthropic's Project Deal infrastructure productized — the pattern will compound. Agent-formed LLCs may go from one to a thousand inside the year.

For a decade, the question "can a piece of software own a company?" was theoretical. As of May 1, 2026, it is operational. The legal system will catch up — but the agent has already moved.

BlockEden.xyz provides enterprise-grade RPC infrastructure across 27+ chains, including the EVM, Solana, Sui, and Aptos rails that AI-agent wallets settle on. As autonomous agents become first-class economic actors, the demands on RPC reliability, rate-limiting, and identity-aware throttling change. Explore our API marketplace to build agent-ready infrastructure on foundations designed to last.

Sources​

Share on Twitter