Evidence review · 12 November 2026 · US entity examples; UK and South African tax-residence examples
What Matters
An agent can execute a treasury rule without becoming the company that owns the treasury. Start with the legal entity, its governing documents and the person or body entitled to delegate authority. Then connect each automated action to that delegation. DN’s decision gates identify the documents a team needs; they do not select a jurisdiction, confer legal personality or predict a tax result.
DN Evidence Block
Reviewed 12 November 2026. The entity comparison covers Wyoming DAO LLC and DUNA rules and Delaware corporation and LLC management rules. Tax examples cover UK company residence, South African company residence and the UK–South Africa treaty. The insolvency example is US Chapter 11. These are separate legal scopes, not a global legal-wrapper ranking. No registry adoption study or operating-speed experiment was performed.
Revision: the former qualitative rankings and wrapper recommendation simulator have been retired. The article now provides a contract-authority decision table and worked operating scenarios. The original DUNA amendment citation was incorrect; the source ledger now points to Wyoming SF0022, enacted for 1 July 2026.
Calling an organization an “agent DAO” describes how some work is performed. It does not tell a supplier who owes an invoice, a contributor who owns delivered code, or a creditor whose assets stand behind a promise. Those questions remain even when the voting and payment process is visible on a blockchain.
The useful comparison is therefore between documented arrangements. A conventional company can delegate bounded execution to software. A DAO can use a recognized wrapper while keeping member decisions onchain. A public governance interface can also sit in front of arrangements that are unclear. Automation is not evidence that one structure is faster, more accountable or better suited to every activity.
What the Named Legal Forms Actually Establish
| Form and jurisdiction | Primary legal anchor | Document to inspect before delegating |
|---|---|---|
| Wyoming DAO LLC | A DAO organized under chapter 17-31 is an LLC. Its articles must describe member management and the extent of algorithmic management. | Filed articles, operating agreement and identified management contracts |
| Wyoming DUNA | A nonprofit association framework with its own purpose, membership and distribution rules | Governing principles, membership basis and permitted payment purpose |
| Delaware corporation | Section 141 generally places management under a board; directors must be natural persons, subject to the statute’s stated governance provisions. | Certificate, bylaws, board authority and delegation |
| Delaware LLC | Section 18-402 provides management and binding-authority defaults that the LLC agreement can alter. | LLC agreement, manager appointment and actual authority |
| A token group with no identified wrapper | This label supplies no entity document for review. | Identify the actual legal arrangement before assuming who contracts or owns assets. |
The sources for those narrow propositions are the current Wyoming Title 17 compilation, Delaware General Corporation Law, board management framework, and Delaware LLC Act, management framework. An incorporation document answers one question. It does not itself establish licensing, tax exemption, sufficient insurance or the enforceability of every transaction.
Wyoming Automation Still Has a Governance Boundary
Wyoming section 17-31-109 places DAO management in members or members together with applicable smart contracts and requires those contracts to be capable of modification or upgrade. The Secretary of State’s Wyoming Secretary of State, DAO LLC guidance describes the formation documents and registered-agent requirement. These are rules for an organization, not a certificate declaring its software a separate legal person.
For an implementation team, map each relevant legal document to the deployed version. A vote approving a policy is not the same record as the transaction that changes a contract. Keep both. When the documents and the software disagree, identify the conflict and obtain a jurisdiction-specific resolution; a dashboard should not silently choose whichever source gives the agent broader powers.
A DUNA is also not simply a company with a fashionable name. Wyoming section 17-32-104 permits profit-making activity for the nonprofit purpose, restricts distributions and allows specified payments, including reasonable compensation. The Wyoming Legislature, DUNA amendments took effect on1 July 2026. Do not turn “nonprofit” into a claim of automatic tax exemption or unrestricted dividends.
The Agent Contract Must Identify the Parties
A proposal to “hire an agent” should first identify what is actually being purchased. Is the organization licensing software, paying a service provider, commissioning deliverables from a company, or engaging a human contractor who operates automation? The model name is not enough to answer that question. Neither is the wallet that receives payment.
DN’s proposed contract schedule has six components: the contracting parties; the accepted deliverable; the permitted tools and spending; evidence of acceptance; change and termination procedures; and ownership or licensing terms. This is a drafting checklist, not a complete contract. The parties must still resolve their applicable law, mandatory obligations and remedies.
For example, a hypothetical DAO wrapper commissions a weekly reconciliation report from a software vendor. The agent may fetch balances and draft exceptions, but the service agreement identifies the vendor and the wrapper, the data it may access and the acceptance test. An onchain payment stream records transfers. It does not prove the report met the contract, transfer every intellectual-property right or decide whether a human worker is an employee.
For the distinct enforceability question, see DN’s agent contracts and accountability analysis. For US copyright boundaries, see the AI-generated work rights analysis only after checking the exact deliverable and contribution evidence.
DN Contract-Authority Decision Gates
| Gate | What the reviewer needs | If unresolved |
|---|---|---|
| 1. Party | Named counterparty and entity evidence where relevant | Clarify who promises performance and receives enforceable rights. |
| 2. Authority | The governing document and delegation covering this action | Escalate the proposed commitment; a wallet signature alone does not settle legal authority. |
| 3. Scope | Amount, assets, counterparties, term and change permissions | Narrow or obtain the missing mandate before enabling that workflow. |
| 4. Acceptance | Deliverable, test, reviewer and dispute path | Do not treat successful code execution as proof of satisfactory performance. |
| 5. Exit | Revocation, data return, key rotation and unfinished obligations | Design the handover before depending on continuous execution. |
| 6. Distress | Asset ownership, payment priorities and escalation owner | Obtain insolvency-specific instructions rather than automatically emptying a wallet. |
There is no aggregate score. A missing party cannot be compensated for by excellent logs. The table is useful because it produces a specific next question, document request or operational restriction. A complete table records the review performed; it does not approve the entity or override counsel, directors, members or a court.
DN Contract-Authority Gates: downloadable documentation templates
The blank record preserves six unresolved areas. The constructed example marks only its first reference as recorded; it describes no actual organization or dispute. A recorded reference is not verified evidence. Use aliases, not personal names, confidential documents or wallet secrets. These files contain no legal approval or numerical verdict.
Schema version 1.0.0. The source review date concerns this article; the template has no record-creation timestamp. For a user-created copy, keep the evidence statuses accurate and record the creation time separately from any evidence review.
party- Who the contracting parties are
authority- The governing document and delegation
scope- Permitted activity and limits
acceptance- Deliverable and acceptance process
exit- Termination, revocation and handover
distress- Asset ownership and distress escalation
Company Residence Does Not Follow the Cloud Region
First identify the taxpayer and the applicable domestic rules. HMRC’s company-residence overview describes UK incorporation and central-management-and-control rules, with exceptions and treaty treatment. It does not say that a company becomes resident wherever an API request executes.
South Africa provides a separate example. SARS Interpretation Note 6, Issue 3, dated 30 June 2023 explains domestic incorporation/formation and place-of-effective-management considerations, including treaty qualifications. Its analysis examines where key management and commercial decisions are actually made, distinguishing these from routine operations. A server-location list cannot answer that factual question by itself.
For an automated business, retain evidence of who sets strategy, approves capital commitments, changes the agent’s mandate and deals with a crisis. Also retain the actual conduct: a nominal board location is not a substitute for the decisions that happened. DN infers from these two official frameworks that runtime location should be treated as an operational fact to examine, not a universal tax-residence switch. Other countries can apply different tests.
Treaty Relief Is Another Gate
The UK–South Africa synthesised treaty text, updated 17 June 2026, illustrates why an old residence shortcut can fail. Its MLI-modified rule for dual-resident persons other than individuals calls for competent-authority agreement considering management, constitution and other factors. Without agreement, treaty relief is limited to what those authorities agree. The authentic treaty and MLI texts prevail over the synthesis.
That is one treaty, not a global rule. Nor does establishing residence settle the classification of a payment, beneficial ownership, permanent-establishment exposure, withholding procedure or all conditions for relief. A vendor invoice paid through an agent still needs those questions considered for the actual payee and income. This article supplies no withholding rate and no “agent tax haven” score.
A Liquidation Function Is Not a Bankruptcy Procedure
A smart contract can sell collateral when a threshold is crossed. That is an execution rule. It cannot by itself establish whose property is being sold, whether the debtor is in a legal proceeding or which claims have priority. The organization’s financial distress and the agent’s lack of gas are also different events.
In the US Chapter 11 example, the federal courts’ Bankruptcy Basics explains the automatic stay and its exceptions. A technical ability to transfer an asset is not permission to ignore an applicable stay. Whether a specific action is covered requires the actual case and legal analysis.
The practical design task is to preserve an asset inventory, identify who can suspend further discretionary commitments, and retain a controlled handover path for the legally authorized decision-maker. Record open orders, collateral, third-party property and unperformed contracts. A “withdraw everything” button can destroy evidence or bypass distinctions the eventual reviewer needs. Avoid describing a system as bankruptcy-ready merely because it can liquidate a portfolio.
Registration and Identity Are Different Records
A machine identity can identify a service endpoint or key. A corporate filing can identify an organization formed under a particular statute. Neither substitutes for the other. In a procurement file, keep the registry reference, legal name, service identifier and authority document in separate fields, with the relationship between them explained.
This is also the limit of the proposed corporate-agent registry theme. The laws reviewed here support particular existing organizational forms. They do not establish a worldwide registry of autonomous legal persons or show that jurisdictions are competing to grant residency to software. New proposals belong in a dated legislative watchlist until their text, enactment and effective scope can be verified.
A Practical Hybrid Governance Stack
Retain the original operational separation: agents monitor, simulate and execute bounded tasks; contracts enforce specified permissions; members or boards approve the matters reserved to them; authorized people handle exceptions; reviewers examine records. The governing law and documents determine which roles actually exist and what duties attach. This is an operating design, not a universal legal allocation.
Test the design with an ordinary failure: the vendor changes a payment address halfway through a task. Who can approve the change, what record proves it, what payment limit applies, and how is a rejected instruction handled? If those questions produce contradictory answers, changing the organization’s label will not resolve the conflict.
Are AI delegates changing governance? Separate power, participation and attribution
Operating-method extension · 12 November 2026. A DAO can observe votes without knowing whether a human, a script or an AI-assisted workflow chose them. Before changing quorum or delegation rules in response to “AI participation,” define the decision being measured and the evidence connecting an actor to an action. Faster voting and more active addresses are not, by themselves, proof that machines displaced human control.
This section complements the legal-authority gates above. The legal record asks who may bind the organization. The participation record asks how a particular decision was reached, how much review it required and what economic incentives might influence it. Neither substitutes for the other.
Count the right unit before drawing a trend
Separate addresses that cast votes, voting power exercised, eligible voting power, unique controllers whose identity is actually established, and delegates with an attributable operating method. Those denominators answer different questions. One controller can use several addresses; one delegate can exercise authority entrusted by many holders; an inactive address can still be represented through a delegate.
Record the proposal, snapshot, voting strategy, delegation rules and the version of the governance configuration. ENS’s documented process provides a concrete example: holders must delegate voting rights before voting or creating proposals, and governance includes different stages and proposal types. That is a reason to inspect the actual process, not to treat every DAO’s displayed turnout number as directly comparable. Primary source: ENS governance process.
For AI attribution, keep separate evidence labels: operator-declared AI assistance; independently reproduced workflow under a stated test; inferred automation; and unknown. A declaration is not independent verification. A reproducible test says what happened in that environment, not how every historical vote was decided. An inference from timing should retain its uncertainty instead of becoming a binary “AI voter” label.
A before-and-after participation chart cannot alone establish causation. Changes in proposal importance, token distribution, delegation, incentives, voting eligibility or the observation window may coincide with a new agent. A defensible study needs the relevant configuration history and a comparison design that addresses those alternatives. This article has not performed that study and establishes no percentage of DAO votes controlled by AI.
Paid participation and paid direction are different incentive designs
Ask what triggers a payment. It may reward attendance, research, maintaining a public rationale, delegating to an active participant, a particular vote direction or a specified outcome. Those arrangements can create different conflicts and review obligations. Do not classify them all as secret vote buying, and do not assume a disclosed program is free of concentration or gaming risks.
ENS’s published EP6.31 temperature-check proposal is marked passed and describes a delegation-incentive pilot. Its text says rewards target sustained activity rather than particular vote choices and specifies caps and time-weighted conditions. This is documentary evidence of that proposed design, not DN verification of implementation, payouts or governance effects. Its status as a passed temperature check is not treated here as proof that every proposed operation has occurred. Primary source: ENS EP6.31.
DN recommends an incentive disclosure record that names the payer, recipient, eligibility, payment trigger, amount or calculation, period, caps and relevant conflicts. Preserve both the stated terms and actual payment evidence where obtained. A treasury transfer without its agreement does not establish what the recipient was paid to do.
For a conditional example, consider two fictional delegates receiving the same amount. One is paid to publish a source-linked analysis regardless of the vote; another receives payment only after supporting a named option. The cash totals are equal but the incentives are not. A policy should address the condition, not just the presence of compensation. Whether a particular arrangement is lawful or permitted depends on the applicable rules and facts, not the label used in a dashboard.
Measure review load before blaming proposal volume for apathy
More proposals can create more work, but counts alone do not measure that work. A short routine renewal, a treasury restructuring and a contract upgrade may need very different expertise and review time. A decline in turnout could reflect delegation or changes in proposal relevance rather than exhaustion. Treat the workload explanation as a hypothesis to investigate.
A useful first planning measure is the reviewer-hours demanded by a stated queue. Divide proposals into classes, record the expected minutes per review and the number of independent reviews required for each class, and compare that demand with available reviewer-hours. Keep the quality standard fixed when comparing two workflows: a faster unchecked summary is not necessarily the same service as a completed review.
Synthetic example: 20 proposals requiring three reviews of 45 minutes each consume 45 reviewer-hours. A further 10 complex proposals requiring two reviews of 90 minutes consume 30 hours. With 60 hours available, the total 75-hour demand exceeds the stated capacity by 15 hours. This is a staffing scenario, not evidence that any DAO’s participation has fallen because of AI-generated proposals.
DN Governance Review Load Calculator
Blank values are unknown, not zero. The example is invented; it is not a product quote or measured outcome.
Not calculated.
Planning arithmetic only. Review durations, quality and capacity are not independently measured. A zero review count is an explicit scenario choice, not an endorsement. No causal turnout, AI participation or quorum recommendation follows. The widget runs locally without wallet access or network requests.
The model does not prescribe a quorum, proposal deposit or approval standard. Those are governance choices with access and accountability consequences. Batching can reduce repeated setup work but also join unrelated decisions. Deposits can deter low-value submissions but also exclude useful contributors. Longer windows can allow review but delay necessary action. Test a change against a stated problem rather than assuming every friction is beneficial.
Staking concentration does not map to the same rights on every network
Separate consensus participation, base-protocol governance and governance of a staking application. Ethereum’s official governance description places core protocol changes in an off-chain process involving multiple stakeholder groups. Holding or operating staked ETH is not equivalent to possessing a universal token-weighted vote that automatically changes Ethereum’s protocol rules. Primary source: Ethereum governance.
Solana’s governance FAQ describes a different, version-specific arrangement: stake-weighted SGP signaling votes and a distinction between a directional decision and detailed SIMD implementation. It also describes staker override of a validator’s vote and labels the linked constitution as proposed. Preserve those qualifications rather than presenting the page as proof that every protocol change is already decided through one immutable voting rule. Primary source: Solana governance FAQ.
A liquid-staking provider’s share of delegated stake can therefore matter for one control surface without proving it controls every governance decision. Record who operates validators, who holds application-governance rights, who can change contracts and what rights the underlying staker retains. A provider brand, validator address and beneficial decision-maker need not identify the same actor.
The actionable question is which specific decision could be influenced through a documented right. Avoid converting a concentration percentage into a universal “governance control” score. A rights map can expose an important dependency even when it cannot estimate the probability of misuse.
DN Governance Participation and Rights Record
Use this evidence record offline. Blank fields remain unknown. The fictional example illustrates how to record gaps; it is not a completed audit or permission to act.
| Review field | Evidence to preserve |
|---|---|
| Decision and window | Identify the proposal, dates, configuration and observation denominator. |
| Operator attribution | Separate declared assistance, reproduced tests, inference and unknown controllers. |
| Voting and delegation rights | Preserve the strategy, snapshot, delegation and override rules. |
| Incentive terms and payments | Record payer, recipient, condition, amount, period, caps and payment evidence. |
| Review demand and capacity | Keep the quality standard, expected minutes, reviewer count and capacity basis. |
| Consensus, protocol and application rights | Map each concentration measure to the specific right it could influence. |
| Unresolved questions and corrections | Record conflicting evidence, responsible reviewer and release or correction conditions. |
No download requested.
A documentary template, not an AI detector, legal opinion, concentration measurement or voting authorization. Unknown controller identities and unverified payment claims must remain unresolved.
What evidence would change a governance policy?
For an attribution concern, require a defined set of decisions and attributable operators, with unknowns retained. For an incentive concern, inspect payment conditions, actual eligible activity and concentration before and after the program. For a capacity concern, collect review-time and completion evidence without counting a rubber-stamp response as an adequate review. For a control concern, identify the exact right and the procedure that makes it effective.
Do not turn the resulting record into a roster of alleged covert AI operators. A classification error could mischaracterize a legitimate delegate. Use the evidence to improve disclosure, review scope or authority boundaries, and provide a correction route for disputed records.
Method and change log: added 12 November 2026 for outlines 102, 104, 105 and 140. This is an operating framework, a capacity calculator and a documentary rights record. No AI-voter attribution dataset, causal turnout study, reward-payout audit or current network-concentration measurement was produced. It neither changes the legal conclusions above nor treats a completed worksheet as approval to vote or execute.
Methodology, Limits and Corrections
This revision replaces unsupported organizational ratings with source-specific legal anchors and DN’s original review gates. It does not measure DAO adoption, compare court outcomes, certify compliance or choose a structure. The checklists and scenarios are editorial methods, not findings about a named business. This review does not recommend a legal provider or rank jurisdictions. It provides information, not individualized legal or tax advice.
12 November 2026 change log: corrected the DUNA source, clarified contracting parties and delegated authority, added scoped residence/treaty and insolvency boundaries, and removed the recommendation simulator. The canonical URL and existing media are retained. Send corrections identifying the jurisdiction, provision, effective date and affected sentence to DN’s corrections desk.
Frequently Asked Questions
Can an AI agent be the director of a Delaware corporation?
Section 141(b) requires directors to be natural persons. A software service used by a board is not itself a qualifying director under that provision.
Does a Wyoming DAO LLC make its agent a legal person?
The reviewed rules concern the LLC and its governance. Algorithmic management does not establish separate legal personality for the software.
Does a DUNA mean no revenue and no paid services?
No. The statute allows profit-making activity for the nonprofit purpose and specified compensation and other payments. The purpose and distribution restrictions still require review.
Can moving a model to another cloud region change the company’s treaty residence?
Do not infer that result. Identify the company, domestic residence tests, actual decision-making and the relevant treaty, including modifications and any required authority determination.
Does an onchain payment prove that a contractor performed?
It proves the recorded transfer. Acceptance, deliverable quality, contractual rights and the identity of the parties need their own evidence.
Which wrapper wins?
No winner is established here. The appropriate arrangement depends on the activity, parties, funding, governance documents and applicable law. The decision gates expose unresolved questions before automation is expanded.




