Relationship managers, paraplanners, portfolio managers and operations staff all fall within Article 4 of the EU AI Act once a wealth manager gives them AI agents to work with. The firm must ensure they have sufficient AI literacy. Because those agents draft advice and stage trades, sufficient means knowing what the agent checked and what the adviser still answers for.
What does Article 4 ask of a wealth manager?
Article 4 requires providers and deployers of AI systems to take measures, to their best extent, to ensure a sufficient level of AI literacy among staff and other people operating or using AI systems on their behalf. The level depends on their experience and training, the context of use, and the people the system affects.
In wealth management the affected people are clients whose savings, pensions and estates depend on decisions that compound over years. A small error in a risk profile, carried into a portfolio, does not show up as a complaint next week. It shows up as an unsuitable allocation discovered at a review, or never. That long feedback loop is why literacy matters more here than the volume of decisions suggests.
The firm also carries suitability duties for investment advice and portfolio management, set under EU financial services law and supervised nationally with ESMA coordinating. Those duties assume the adviser understood the recommendation. An adviser who signs off an agent-drafted suitability report without knowing what the agent read has not met them.
Who in the firm does the duty cover?
- Relationship managers and advisers who present recommendations an agent drafted.
- Paraplanners who use agents to assemble fact-finds, cash flow models and suitability reports.
- Portfolio managers whose rebalancing agent proposes trades across many client accounts.
- Client onboarding and operations who release agent-prepared account openings and transfers.
- Compliance and file checkers who review advice files and must judge the agent's part in them.
- Introducers and outsourced paraplanning firms working on your clients for you.
Outsourced paraplanning is common in the sector and often missed. Article 4 covers people acting on your behalf, so the firm must know what those paraplanners understand about your agents.
What must advisers and portfolio teams understand about agents?
The detail in what AI literacy means for wealth managers using AI agents is worth reading in full. The core is six points.
- Depth by capability. A research agent reads and summarises at depth 1. A suitability agent drafts the report at depth 2 and an adviser sends it. A rebalancing agent may place routine trades at depth 3 inside model-portfolio bounds, while anything that moves client money out, such as a withdrawal or transfer, is depth 4 and held for a named person.
- What the fact-find rests on. Client statements about income, assets and attitude to risk are declared. Custody holdings are sealed. A valuation emailed by a third party is external. Advisers must see which is which before they rely on a figure.
- Where suitability judgement sits. The agent can match a profile to a model portfolio. It cannot see that a client sounded unsure on the phone, or that a family event changed their horizon. Advisers must be trained to add what the record does not hold.
- Vulnerability. Older clients, bereaved clients and clients under pressure from relatives need a person to notice. Agents can flag patterns; advisers must act on them.
- Bulk effects. A rebalancing agent applying one error across hundreds of portfolios is a different risk from one adviser's mistake. Portfolio teams must read held trade batches as a whole, not line by line.
- What the record shows. Each verdict is filed, so compliance can reconstruct who approved what. Advisers should know that their release of a held recommendation is recorded against their name.
How can a wealth manager meet Article 4?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Literacy standard per role | Standards for advisers, paraplanners, portfolio, operations and compliance | Approved policy with an owner |
| Context of use | Training on each agent's depth, sources and held actions | Agent-specific assessment results |
| Client impact | Module on vulnerability, suitability limits and client understanding | Module records; sampled advice files |
| Competent release of held actions | Only trained, current staff can release held transfers and recommendations | Release log with approver and training status |
| Outsourced paraplanners | Contract clause on literacy and attestation | Attestations; file samples |
| Evidence of effect | Seeded advice files with a planted unsuitable recommendation | Detection results per adviser |
| Currency | Refresh when an agent's brief or depth changes | Change-linked training records |
The full method for building this is in how to set up an AI literacy programme under the EU AI Act. Payments firms face a faster version of the same problem; Article 4 in payments shows the contrast.
What will a supervisor or auditor ask?
- Which parts of the advice process do agents perform?
- Ask an adviser: what did the agent check in this file, and what did you check?
- Show us the training record for whoever released this transfer.
- How do you train outsourced paraplanners on your agents?
- How do you detect an error repeated across many portfolios?
- What happened to training after the suitability agent was changed?
What does Article 4 not cover for wealth managers?
It does not set the content or length of training. It does not replace suitability, conduct or client asset rules, which apply in full whether or not an agent drafted the work. It does not cover data protection, which governs profiling clients under GDPR separately. And it does not by itself make any wealth use high-risk under the Act. Whether the Act requires AI literacy training in a formal sense matters less than whether you can show your advisers understand the agents. The guide to AI literacy and certification sets out the general approach.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where the Academy fits
Pryme Intelligence Academy trains people on the real product. Advisers and paraplanners who release held recommendations take the Operator track, with module tests set in wealth management and their own role. Portfolio teams who design agents take the Builder track, which ends with a capstone agent certified in Attest. A firm can choose to tie release rights for each depth to a current certificate, so only certified staff approve held transfers.
Questions people ask
Does an adviser who only reads agent research need training?
Yes, at a lighter level. They act on what they read, so they need to know which sources the agent used and how to check a claim before passing it to a client.
Should compliance staff be trained differently from advisers?
Yes. Compliance staff review files after the fact. They need to read the record of verdicts and judge whether the adviser's review was real, which is a different skill from releasing work.
Do discretionary trades need a person if the agent stays within the model?
Routine rebalancing within model bounds can run at depth 3. Withdrawals, transfers out and anything outside the model should be held for a named person.