AI Portfolio Advice: The US Adviser Duties Behind the Automation

Decentralised News · Editorial by Heath Muchena · Evidence reviewed 11 November 2026. Original DN analysis and local instrument; field testing is claimed only where expressly reported.

“The agent manages the portfolio” describes a technical function. It does not tell the client which entity supplies the service, what the agreement covers or who reviews conflicts in the business model. Those questions become more consequential when a system can move from analysis to recommendations and then to discretionary actions without a visible handoff.

The practical starting point is the relationship. Identify the entity, the client, the agreed service and the applicable legal assessment before examining a model’s performance chart. A technical vendor, an adviser, a broker and a client can occupy different roles in the same architecture. The label “agent” does not settle those distinctions.

What the US adviser sources actually establish

The SEC’s 2019 Commission interpretation describes an investment adviser’s fiduciary duty through care and loyalty, including acting in the client’s best interest. The application depends on the adviser-client relationship and its scope. The interpretation addresses matters including investment advice, execution and monitoring, as well as conflicts and disclosure. It is a framework for the adviser relationship, not a general declaration that software itself is a fiduciary. Source: Commission interpretation IA-5248.

The SEC’s 2017 robo-adviser material is staff guidance, not a newly enacted AI rule. It discusses disclosures, obtaining client information and compliance programs for automated advice. Its examples recognize different degrees of human involvement. Applying it to a modern agent requires examining the actual service rather than assuming that all automated systems operate identically. Source: SEC staff guidance 2017-02.

For readers checking a service, the SEC’s accompanying investor guidance directs attention to the adviser relationship, available human interaction and registration research. A polished interface or a familiar model brand should not replace identification of the entity actually providing advice. Source: SEC robo-adviser guidance announcement.

Claim versus evidence: what the automation claim leaves out

These questions are DN’s evidence method. They are not a checklist that substitutes for the governing standard. A reference can be available yet inadequate, outdated or inconsistent with what the service does.

A 2026 case shows why the business model belongs in the record

On 23 March 2026, the SEC announced a settled order involving Ally Invest Advisors’ cash-enhanced robo-adviser accounts. The SEC said the firm failed to fully disclose material facts and a conflict related to a 30% cash allocation, selected partly to replace revenue forgone from advisory fees. Ally agreed to a $500,000 penalty without admitting or denying the findings. This was a disclosure case involving a named adviser, not proof about every AI agent. Source: SEC proceeding summary.

The operational lesson DN draws is to inspect how the service earns money alongside how it selects investments. An optimization objective and a revenue arrangement can interact. The evidence review should therefore include compensation and conflicts even when the visible interface emphasizes low fees, automation or personalization.

Do not generalize the case into a rule that a particular cash allocation is always unsuitable. The relevant claim is about the disclosed facts and conflict in that proceeding. A review of another service needs its own client relationship, method and evidence.

Trace the client mandate into the system’s actual behavior

Begin with the agreed scope: research support, client recommendations or discretionary management. Record the underlying assessment rather than treating the selected label as a legal conclusion. A system described as research support may be used differently in practice; the review needs to follow actual inputs, outputs and decisions.

The client-profile record should identify which information the service uses, how it resolves inconsistent answers and when it asks for clarification. Do not paste personal financial information into a general worksheet. Reference the controlled profile or assessment by an alias. The useful question is whether the relevant process exists and can be inspected by the responsible reviewer.

The SEC staff’s care-obligations bulletin discusses understanding the investment’s risks, rewards and costs, the investor’s profile and reasonably available alternatives. It also cautions against treating the cheapest option as automatically sufficient. This is staff guidance spanning adviser and broker contexts; the underlying regimes remain distinct. Source: SEC care-obligations bulletin.

DN’s practical extension is to connect those records to the specific system version. If a portfolio recommendation changes after a model, data feed or prompt update, the reviewer should be able to identify what changed. An evaluation conducted against an earlier configuration does not explain the new behavior by itself.

Make oversight concrete enough to inspect

“Human in the loop” is incomplete unless the record says where the person intervenes. A reviewer might approve a strategy before deployment, inspect a sample of recommendations, respond to exceptions or authorize each transaction. Those designs have different operational consequences. Name the role and trigger in the controlled record without implying that one arrangement guarantees compliance.

Document what the reviewer can see. A final recommendation without its data assumptions or intended action may be difficult to assess. Conversely, a large log dump can conceal the material decision. A useful review packet identifies the client mandate, proposed action, relevant constraints and reasons for escalation in a form the reviewer can actually use.

Also record what happens when intervention fails. Who can suspend new recommendations? Who can revoke execution authority? Which open orders or external actions require separate attention? These are design questions for the actual arrangement; the DN tool does not connect to an account or test a kill switch.

Disclosures should describe the service the client receives. If the system produces portfolio suggestions but does not consider every part of the client’s financial situation, a review should examine whether the presentation creates a broader impression. This question follows the staff guidance’s focus on scope and understandable disclosure; it is not answered by attaching a generic AI disclaimer.

Use the DN US Adviser AI Control Record

The instrument records seven areas: relationship, client mandate, system validation, investment assessment, conflicts, disclosure and oversight. Each receives a reference alias, an explicit issue or an unknown. The tool does not read the referenced material, evaluate the portfolio or determine whether a duty has been satisfied.

Select whether the US adviser relationship has been assessed and the proposed automation scope. If the legal relationship remains unknown or belongs to another regime, the result preserves that limitation. Selecting discretionary actions adds questions about execution authority, limits, monitoring and intervention. The selection does not grant permission to execute anything.

Constructed example: a team records references for the relationship, client mandate, model validation and disclosures. It flags unresolved compensation treatment and leaves investment assessment and oversight undocumented. The result is four referenced areas, one explicit issue and two unknowns. It is not a 57% compliance result. The unresolved compensation issue may be material regardless of the other references.

The JSON export contains the selected scope, reference aliases, counts and unresolved questions. Its creation time is a record timestamp, not an audit date. Approval remains null. The tool asks for no client names, balances, account identifiers or private agreements, and it does not send the record to a provider.

Use the record when the service changes

Define review triggers that correspond to the system’s behavior. Examples include a new investment universe, a changed compensation arrangement, a revised client questionnaire, a new external execution tool or a material model update. These are DN’s suggested operational prompts, not an exhaustive list of legal requirements.

For each trigger, preserve the previous version and identify the change being reviewed. A model version alone may be too narrow if the data source, tool permissions or portfolio constraints changed independently. The evidence packet should explain the configuration that produced the relevant recommendation or action.

Keep unresolved items visible to the person making the deployment decision. A deadline should not silently convert “not reviewed” into “no issue found.” If the review scope was limited, record the limitation. An agent can help gather and compare evidence, but a generated summary cannot attest that an unperformed review occurred.

DN’s agent rights and responsibilities article examines broader accountability questions. Its organizational-model comparison addresses entity design. This record is narrower: the documentary chain for an identified US adviser relationship using automation.

Methodology and limitations

The article distinguishes a Commission interpretation, staff guidance and a settled enforcement proceeding. None is treated as a universal law of machine personhood. No claim is made that every portfolio tool, developer, DAO or token arrangement falls under the same adviser framework. Entity status, asset status, exemptions, other regulators and other jurisdictions require their own assessment.

DN’s original tool counts documentation states across seven areas without weighting them or predicting legal outcomes. It does not measure model accuracy, fiduciary performance, investment return or client suitability. A complete record can support a better review while still containing evidence that leads the reviewer to reject a proposed service or require changes.

Frequently Asked Questions

Does using AI make the developer a fiduciary?

The technology label alone does not determine the legal relationship. Identify the actor, service and governing assessment before assigning duties.

Can a disclaimer resolve every conflict?

The record should identify the actual conflict and the applicable treatment. A generic disclaimer does not establish that the relevant facts were appropriately addressed for the relationship.

Does a human review label prove effective supervision?

No. Record the reviewer’s role, information, timing and intervention authority, then inspect how the process works in the specific service.

Does the worksheet assess an investment recommendation?

No. It organizes references and unknowns. It does not recommend an investment, calculate suitability or decide that an adviser has complied with its duties.

Corrections and change log

Send corrections with the official source, legal scope and affected passage to press@decentralised.news. Do not send private client records. 11 November 2026, version 1.0: introduced the US adviser evidence record and distinguished Commission interpretation, staff guidance and the dated Ally settlement. No global fiduciary-duty allocation or compliance ranking is published.