Decentralised News · 2027 planning series · Reviewed 21 November 2026
What Matters
An AI agent can help create and deploy a token, but the launch still has a chain of authority. Someone funds the signer, chooses its permissions, controls the deployed contract, receives the initial supply and decides what can happen to treasury assets and liquidity. The useful question is not whether a token was “made by AI.” It is which powers survived the launch, who can exercise them now, and what evidence supports that answer.
The Signal: Evidence reviewed
Reviewed: 21 November 2026. Documented evidence: Coinbase AgentKit token-deployment capabilities; the ERC-20 interface; OpenZeppelin access and proxy controls; Solana token authorities; Uniswap liquidity mechanics; and an EU public-offer example. DN contribution: a proposed seven-authority record and a fictional handover exercise. DN has not launched a token, tested a live project’s controls or measured a protocol’s security for this article.
DN Alpha Thesis: deployment is the beginning of the control record
The transaction that creates a token is a useful timestamp. It is an incomplete explanation of the business. A launch can move a project from one set of controllers to another: a temporary deployment key, a treasury multisig, an upgrade administrator, a liquidity manager and an operating agent may all appear at different stages. A slogan such as “community owned” compresses those roles into a single claim and can hide the powers that matter most.
DN’s proposed approach is to publish the authority record alongside the launch record. Each material power should have a named technical holder, an accountable operator, evidence at a stated chain height, and a documented change or removal path. That gives researchers a concrete object to recheck after a migration, administrator change or operational incident. It also makes a legitimate project easier to explain without pretending that a checklist predicts token returns.
DN Token Launch Authority Record
Change the illustrative assumptions below. This worksheet runs locally in your browser. It records your declarations; it does not verify evidence, connect to accounts, submit instructions or grant approval.
The initial numbers are synthetic. Unknown evidence remains unresolved. Downloads contain the values you enter; use aliases and avoid private or identifying information.
What does it mean for an agent to launch a token?
There are at least four different claims hiding inside the phrase. An agent may suggest the token’s name and distribution; generate code; call a deployment tool after approval; or operate a wallet that can initiate deployment within previously granted permissions. Those are different levels of discretion. A launch report should state which occurred and preserve the relevant approval and execution evidence.
Coinbase’s AgentKit documentation describes an onchain action framework and a deploy_token action for creating an ERC-20 token with supplied parameters. That establishes a software capability. It does not establish that every AgentKit installation exposes the action, that a particular agent used it, or that deployment transferred every subsequent administrative power to the same wallet.
For a specific launch, ask for the chain and deployed address, transaction receipt, code or template version, resolved parameters, signer address and approval reference. Keep the proposal separate from the committed transaction. If a model drafted one supply figure and the signed deployment used another, the final receipt is the execution evidence; the difference still needs an explanation.
The seven authorities to map
The following is an original DN review framework, not a token standard or a universal security certification. Some rows will not apply to a particular design. A reviewer should document why, instead of treating a blank field as proof that a power does not exist.
| Authority | Question the record must answer | Evidence to request |
|---|---|---|
| 1. Deployment and signing | Who could authorize the creation transaction, and who can still use that signer? | Receipt, signer policy, approval reference and credential lifecycle. |
| 2. Supply | Can anyone create additional units or change the rules governing supply? | Deployed code, applicable role state, supply rules and change history. |
| 3. Implementation | Can the program or contract logic be replaced or redirected? | Implementation mapping, upgrade controller and applicable delay or approval process. |
| 4. Transfers and restrictions | Can a controller pause, freeze, restrict or otherwise alter transfer behavior? | Actual functions or extensions, controller and conditions for use. |
| 5. Treasury | Who can move launch proceeds and operating reserves? | Treasury address, signing configuration, policies and any delegated spending rights. |
| 6. Liquidity | Who can withdraw or alter the liquidity described in the launch materials? | Pool and position identifiers, ownership, permissions and lock terms where claimed. |
| 7. Distribution and claims | Who can change allocations, releases, eligibility or the receipt of fees? | Distribution contracts, vesting configuration, beneficiary records and amendment powers. |
Each row needs both a technical and an organizational answer. “A multisig controls the treasury” describes a mechanism. It does not identify whether the signers are independent, who can replace them, or whether another route permits spending without their ordinary approvals. Those questions belong in the record whenever they materially affect the claim being made.
Resolve the deployed design before applying a template
The ERC-20 standard specifies a fungible-token interface, including transfers, balances and allowances. It does not require one universal ownership, minting, upgrade or revenue-distribution design. Consequently, “ERC-20 token” is not a complete answer to the supply, control or holder-rights questions in the table.
OpenZeppelin documents both a simple owner model and granular roles. An owner can be a contract, and administrative roles may control other roles. For a review, follow the control chain rather than stopping at the first owner address. Record the applicable design and version; do not infer the deployed permissions merely because a project says it used an established library.
Upgradeable designs require an additional distinction. The address users interact with may delegate behavior to an implementation, and the upgrade mechanism has its own authorization path. OpenZeppelin’s proxy documentation illustrates these relationships. DN’s practical inference is that a “fixed supply” assertion needs to be read together with any mechanism that can change the code enforcing that supply.
On Solana, the vocabulary changes. The official authority documentation distinguishes mint and freeze authorities from token-account ownership and other authority types. Setting a supported authority to None removes that selected role; additional token extensions can introduce additional controls. A reviewer should therefore identify the program and enabled extensions before importing an Ethereum-style owner checklist.
A removed owner is a narrow claim
One control being removed does not explain the entire system. Suppose a project presents a transaction showing that a particular owner role was relinquished. The appropriate review question is: which functions did that role govern? The next question is whether supply, implementation, distribution, treasury and liquidity have separate controllers. The evidence may support the narrow statement while leaving the broader decentralization claim unresolved.
Use a claim-to-evidence table with four columns: the exact claim, the relevant component, the state observed and the remaining uncertainty. This prevents a genuine technical milestone from being stretched into a promise it does not establish. It also avoids the opposite error of treating every retained administrative power as proof of fraud. Some systems deliberately retain emergency controls; readers need to know their scope and governance.
Liquidity is a separate authority surface
Token administration and market liquidity should have different rows. In Uniswap v2’s pool model, liquidity providers receive tokens representing their share of a pool. Uniswap’s v3 documentation separately describes reducing a liquidity position. These examples show why the asset contract alone cannot answer who can withdraw the liquidity advertised to buyers.
For a claimed lock, ask what precisely is locked: a position, a representation of a position, a subset of liquidity, or the ability of one controller to perform one operation. Record the lock contract, beneficiary, expiry, withdrawal conditions and any relevant administrative route. A screenshot of a countdown is not enough to reproduce the analysis. Nor does a lock establish that market depth is adequate for a particular order size.
The useful commercial distinction is between a verifiable control and a market outcome. A distribution rule can be documented; the market price after distribution cannot be guaranteed. A published liquidity position can be inspected; future exit liquidity still depends on market conditions and the position’s behavior. Keep those two kinds of statement separate in launch communications.
Design the agent’s permission boundary before the launch
DN proposes separating preparation, approval, execution and reconciliation into distinct responsibilities. Preparation can assemble the intended parameters and evidence. Approval should identify the exact proposed action. Execution should enforce the permitted method, network and values. Reconciliation should compare the resulting state with the approved state and record any discrepancy.
The key operating question is what the agent can do after the initial deployment. An agent allowed to create a test token does not automatically need treasury spending, upgrade control or liquidity-management authority. If one workflow requires several powers, record why they are combined, what limits apply and who can stop the workflow. The control should be evaluated against the actual permitted calls, not the friendly name shown in the chat interface.
Limit testing to systems and environments the operator is authorized to test. A useful handover exercise can be performed with a mock or test deployment: the new controller demonstrates its intended authority, and the retired credential fails to perform the removed action. Label those results as test-environment observations. They are evidence about the tested arrangement, not proof of a different mainnet deployment.
Worked example: a successful launch with an incomplete handover
Fictional scenario: a team approves an agent to deploy a token with an initial supply of one million units. The deployment receipt matches the approved parameters. The team then publishes “the AI launched a community-owned token and the launch wallet has been retired.” No real token or project is represented by this example.
The review resolves six of the seven authority rows. Supply administration is documented; the applicable implementation mapping is captured; transfer controls are described; treasury and distribution arrangements have named controllers; and the deployment wallet’s credential has been disabled. The liquidity row contains only a marketing screenshot and an unidentified position owner.
The result is an incomplete authority record. Six resolved rows do not establish who can change the seventh. DN’s proposed tool therefore reports the unresolved liquidity row instead of averaging the record into a high safety score. The correct publication language is narrower: the creation transaction and specified handover steps were documented, while liquidity control remains unverified.
Now change the example. The team identifies the liquidity position and its controller but also discovers that the operating agent still has a separate treasury delegation. The record becomes more complete, but the claimed retirement of all agent authority conflicts with the observed permission. Completeness and consistency are different properties. The conflict requires an explicit explanation and a new handover test, not a better percentage.
If that delegation is intentionally retained, the team can disclose its purpose, expiry and spending boundary. If it is unintended, the authorized operator can remove it and preserve the subsequent verification evidence. Either way, the reviewer has turned a vague decentralization claim into a falsifiable statement about a particular power at a particular time.
Use the DN Token Launch Authority Record
The accompanying instrument is a documentation aid based on the seven rows above. For each row, select whether the authority is evidenced, unresolved, inconsistent with the public claim, or not applicable with supporting evidence. Record only public references or internal record identifiers. Private keys, recovery phrases and credential values have no place in the worksheet.
The instrument gives conflicts priority. Any inconsistent row produces a claim-conflict result. Otherwise, unresolved rows produce an incomplete-record result. Only when all seven rows are evidenced or supported as not applicable does it report that the authority record is documented. That wording deliberately stops short of a security verdict, regulatory conclusion or investment recommendation.
Use the result to assign the next research task: find the implementation controller, obtain the distribution configuration, or clarify a liquidity claim. Keep the reviewed chain height or time in the evidence packet. A documented result can become stale after an upgrade, role transfer, new delegation or change to the operating configuration.
The legal actor does not disappear inside the software
Technical automation and legal classification are separate questions. For a concrete example, MiCA Article 4 places conditions on certain public offers of crypto-assets in the EU and contains specific exemptions and limits. That is a reason to determine the activity, asset, people and jurisdictions involved. It is not a basis for assuming that every token launch follows the same rules.
Before a public offering or promotional campaign, the responsible team should obtain a scoped legal assessment and retain the underlying facts. Useful questions include who makes the offer, who controls proceeds, what purchasers receive, where the communications reach people, and whether intermediaries provide regulated services. Calling the deployer an agent does not answer those questions. This article does not determine securities status, licensing duties or liability in a particular case.
What a useful public launch record looks like
A concise public record should start with the network, asset identifier, deployment receipt, stated purpose and versioned authority table. Link the technical evidence supporting important claims and identify the review date. Where information is deliberately withheld, explain the boundary without publishing sensitive credentials or personal security details.
Use a change log that records changes to claims as well as contracts. If a treasury delegation is extended, or a distribution policy changes, a reader should be able to see which version of the description applied when they acted. A token’s name, symbol and social account are useful labels; the durable research references are the identifiers, states, permissions and records behind them.
For broader investor screening, use DN’s ten evidence checks for AI-agent token projects. For economic claims, use the token-holder payout-rights reconciliation. The authority record answers who can change the system; those studies answer whether the product evidence and holder economics support the narrative.
Methodology and limitations
This is documentation-based analysis and an original proposed review workflow. The seven authorities are editorial categories, not seven mandatory fields imposed by one blockchain standard. The example is synthetic. No contract audit, launch transaction, credential revocation, liquidity withdrawal or attack was executed for this research.
Public records may not reveal common control, undisclosed agreements, compromised devices or future actions. Complete documentation can describe a risky design, and well-governed controls can still fail. Readers should use qualified technical and legal review for consequential decisions and replace illustrative states with evidence from the exact deployment being assessed.
Editorial scope: this article evaluates authority records and makes no token recommendation. The instrument is a local planning aid and does not connect to a wallet or submit transactions.
Sources and evidence register
Primary sources reviewed on 21 November 2026. The article distinguishes documented facts, DN analysis and synthetic examples. Recheck changeable rules and configurations before acting.
- Coinbase AgentKit reference — Documented framework and deploy_token capability; no live deployment claimed.
- ERC-20 token standard — Required token interface; no universal administrative or economic rights model.
- OpenZeppelin Access Control — Ownable, role-based control and administrative roles.
- OpenZeppelin Proxy API — Implementation and upgrade-control relationships.
- Solana Set Authority — Distinct authority types and role revocation; extensions may add controls.
- Uniswap v2 Pools — Liquidity-token claims on a pool.
- Uniswap v3 Decrease Liquidity — Separate liquidity-position operations.
- ESMA MiCA Article 4 — Scoped public-offer example with conditions and exemptions; not universal legal advice.
Frequently asked questions
Can an AI agent create a cryptocurrency token?
Software agents can be connected to token-deployment tools. Whether a particular agent can create a token depends on its configured actions, wallet permissions and approval process. A documented capability is not evidence that a particular launch occurred.
Does renouncing token ownership remove every control?
It removes the power governed by that specific ownership mechanism. Other supply, upgrade, treasury, liquidity or distribution authorities may be separate and require their own evidence.
Does an agent-created token automatically pay holders the agent’s revenue?
No universal token interface establishes that economic promise. Review the documented holder mechanism, distribution terms and actual payment evidence separately.
Is the DN authority record a security audit?
No. It organizes documentation and identifies unresolved or conflicting control claims. It does not test smart-contract security, predict returns or certify regulatory compliance.
What should be checked after the launch wallet is retired?
Check the exact powers claimed to have been removed, the new controller’s intended authority and any separate delegations, roles or recovery routes. Preserve the environment, time and result of each authorized verification.