AI for Asset Management: Research, Memos, RFPs, Reporting and Operations Under SEC, EU and FCA Rules

AI for asset management, from front office to back office

Asset managers already run plenty of quantitative models. What has changed is the arrival of models that read and write, which reach the parts of the firm that work in documents rather than numbers. The table maps the main functions to the AI tasks that suit them.

Private markets firms face a different set of documents, mostly data rooms and portfolio company accounts; our guide to AI across the private equity deal cycle covers that work. This article is about long-only, multi-asset and hedge fund managers investing mainly in listed markets.

Research synthesis over filings, transcripts and broker notes

An analyst covering 40 companies reads several hundred documents a quarter. AI helps most with three kinds of research work.

First reads. A structured summary of a 10-K, an interim report or an earnings call: segment results, guidance changes, capital allocation and the questions analysts pressed on. The summary should quote and cite the passage behind each point.

Change detection. Comparing this quarter’s call with the last four and listing what management stopped saying, started saying or hedged. Language models are better at this kind of comparison than at forecasting, and it surfaces questions for the next management meeting.

Thematic screens. Searching a coverage universe’s filings for exposure to a theme, such as a tariff, a regulation or a supplier, and returning a list of companies with the relevant passages.

Licences limit what you can send to a model

Public filings carry no licensing restriction. Most other research inputs do. Broker research, consensus estimates, ratings, index data and exchange or vendor market data typically arrive under agreements that limit redistribution and say how the content may be stored and used. Before indexing any of it in an AI tool, check each agreement for clauses on redistribution, storage, derived works and automated or machine processing, and ask the provider when the terms are unclear. Many terminals and research platforms now offer their own AI search over their content, which keeps the processing inside the provider’s licence. Running a model on infrastructure you control does not remove licence limits, but it does remove the question of whether sending the content to a third-party AI provider counts as redistribution.

Expert networks and MNPI

Transcripts and notes from expert network calls are useful research input and a known MNPI risk. If they go into a searchable AI index, the index needs the same access controls, pre-clearance records and review as the transcripts themselves.

AI for portfolio managers: memos and briefings

Portfolio managers want less to read, not more. Two uses earn their place quickly.

Investment memo drafts. When an analyst recommends a new position, the memo follows a house template: thesis, valuation, catalysts, risks, position size and exit conditions. A model can draft it from the analyst’s notes, the model output and the cited research, leaving the analyst to sharpen the argument. Every figure should come from the valuation model or a cited source, and the analyst should own the final text.

Morning briefings. A short daily note per portfolio covering overnight news on holdings, earnings due, rating changes and any position that moved beyond a set threshold. An agent can assemble it from news feeds the firm is licensed to process, the order management system’s positions and the risk system’s limits, and flag only what changed.

For desks that want analytics alongside the text, the trading desk decision support playbook shows how a governed assistant sits beside traders without placing orders, and the financial statement analysis use case covers ratio and trend analysis with citations.

RFPs, DDQs and consultant questionnaires

Institutional sales teams spend weeks a year on requests for proposal, due diligence questionnaires and investment consultant databases. Most questions have been answered before, in a previous RFP, the compliance manual, the fund prospectus or the ESG policy. The work is finding the latest approved answer and adapting it.

Retrieval over an approved content library does that well. For each question, the model drafts an answer from the most recent approved source, cites it, and marks questions where no approved answer exists or where the source is older than the firm’s review cycle. Performance figures and AUM should be pulled from the official data set, never written by the model. Compliance still reviews before submission, because RFP responses can be advertisements under the marketing rule.

The RFP response agent is built on this approach, with a reviewer approving each response before it leaves the firm.

Client and fund reporting

Quarterly commentary, factsheets and client reports combine performance and attribution data with narrative. A model can turn attribution output into a first draft of the commentary, in the right tone for retail factsheets or institutional reports, and check that the narrative matches the numbers: if the text says stock selection in healthcare helped, the attribution should show it.

Client service teams also field repeated questions about fees, share classes, dealing deadlines and fund documents. An assistant that answers only from approved documents, as in the investor relations assistant use case, can draft responses for the team and escalate anything outside its sources.

Operations: trade breaks, reconciliations and the shorter settlement cycle

Operations teams work through exception queues every day: unmatched trades, cash and position breaks between the accounting system and custodians, failed settlements and corporate action differences. AI does not fix a break, but it can read the trade record, the custodian file and the broker confirmation, identify the likely cause, such as a wrong settlement instruction, a price or quantity mismatch, or a missing allocation, and draft the explanation and the query to the counterparty.

Settlement timetables make this work more pressing. The SEC’s move to T+1 for most broker-dealer transactions took effect on 28 May 2024, and the same rules require registered investment advisers to keep records of allocations, confirmations and affirmations for institutional trades. The EU plans to move to T+1 on 11 October 2027, with ESMA setting interim deadlines before then. Less time to resolve a break means more value from a tool that explains it quickly. The reconciliation agent shows the exception-handling pattern with a human closing each item.

Compliance: MNPI, Section 204A and the code of ethics

An AI assistant that can search research, holdings, orders and communications is a new place where sensitive information collects. Two US obligations make that a compliance design question, not just an IT one.

Section 204A of the Investment Advisers Act requires covered advisers to establish, maintain and enforce written policies and procedures reasonably designed to prevent misuse of material nonpublic information. If expert call notes, issuer meeting notes or private-side information sit in an AI index, the policies need to cover who can retrieve them, whether the index respects information barriers, and how queries are logged.

Rule 204A-1 requires a code of ethics that includes holdings and transaction reporting by access persons and pre-approval of IPO and limited offering investments. A firm that lets an assistant search the order book or upcoming trades has to decide who counts as an access person for that tool.

AI also helps compliance teams directly. Typical uses include triaging communication surveillance alerts with a short explanation of why each was raised, pre-reviewing marketing material against the marketing rule, and checking personal account dealing requests against restricted lists. Each still needs a reviewer who closes the item.

AI in asset management regulation: SEC, EU and UK

United States

The SEC regulates AI at asset managers through existing rules. In June 2025 it withdrew the 2023 predictive data analytics proposal, which would have imposed specific conflict rules on advisers’ use of such technologies.

The Division of Examinations’ fiscal 2026 priorities name automated investment tools, AI technologies and trading algorithms or platforms as an area of focus. Examiners will check whether firms’ representations about AI are accurate and whether policies and procedures exist to monitor or supervise AI use in fraud prevention, back-office operations, anti-money laundering and trading functions. Cybersecurity reviews will include the controls firms use against new risks from AI.

The marketing rule, Rule 206(4)-1, applies to any AI claim in a pitch book, RFP or website. The SEC’s 2024 settlements with Delphia and Global Predictions, $400,000 in combined penalties for misstated AI capabilities, remain the reference case. Amended Regulation S-P, with compliance dates of 3 December 2025 for larger firms and 3 June 2026 for smaller ones, requires oversight of service providers holding customer information, which includes AI vendors used for client reporting or service.

In a February 2026 speech to the Investment Company Institute’s winter board meeting, the director of the SEC’s Division of Investment Management said liability worries are the main brake on adoption and invited firms to approach staff for guidance or no-action relief. He stressed that his remarks did not necessarily reflect the Commission’s views.

European Union

ESMA’s public statement of 30 May 2024 gave initial guidance to firms using AI when providing investment services to retail clients. It said firms must meet their existing MiFID II obligations on organisation, conduct of business and acting in the client’s best interest, and named risks such as algorithmic bias, poor data quality, opaque decision-making and overreliance on AI.

The fund directives do not mention AI, but two recent changes matter for AI vendors. Directive (EU) 2024/927 amended AIFMD and the UCITS Directive, including the delegation rules, with national measures for most provisions applying from 16 April 2026; a manager still cannot delegate so much that it becomes a letter-box entity, which is worth remembering when an outside AI service starts doing core portfolio or risk work. DORA has applied since 17 January 2025 and covers alternative investment fund managers other than sub-threshold managers, along with investment firms and other financial entities, so an outside AI provider will usually be an ICT third-party service provider subject to the firm’s DORA risk management and contract requirements.

United Kingdom

The FCA does not plan AI-specific regulation, as it said when launching the Mills Review in January 2026. Its October 2026 good practice publication on AI in financial services asks firms for clear senior accountability, outcome testing, audit trails and exit plans for AI vendors, and warns against relying on untested vendor assurances. UK managers whose funds reach retail investors also apply the Consumer Duty to AI-written fund communications.

AI for hedge funds: what is different

Hedge fund managers raise two issues that long-only firms meet less often.

Systematic and discretionary are different problems. Systematic funds have long used statistical and machine learning models on price, fundamental and alternative data. Those models are validated, backtested and monitored as trading systems. Language models add a research layer for discretionary managers and a data-processing layer for quants, for example extracting signals from text. The SEC’s examination priorities explicitly cover trading algorithms and AI in trading functions, so model governance documentation matters for both.

Information is the edge, and the risk. Expert calls, channel checks and alternative data are core inputs, and each carries MNPI, privacy or licensing questions. An AI tool that makes this material easier to search also makes it easier to misuse. Access controls by strategy and team, logging of every query, and compliance review of how new data sets enter the index are the minimum.

This section is about how hedge fund managers use AI inside the firm. It does not assess funds marketed as AI-driven, which raise marketing rule questions of their own.

Which asset management workflows can go to a cloud model

The table assumes enterprise terms that exclude training on customer data. It is a starting point for the firm’s own data classification.

How VDF AI fits an asset manager

VDF AI runs inside the manager’s own infrastructure: its data centre, a private cloud account or a segregated network with no outbound access. In local-only mode, research notes, holdings, order data, client reports and the logs of every query stay in that environment, which also keeps licensed content off third-party AI services. VDF AI carries no regulatory approval or certification of its own; it is software that operates within the manager’s compliance program, DORA or vendor oversight framework and information barrier policies.

Each research team, strategy or client book can have its own knowledge collection, and retrieval returns only passages from documents the user is permitted to open, which lets compliance mirror the firm’s wall-crossing and access person rules inside the assistant. Governed agents carry out the repeatable tasks in this guide, from RFP drafting and fund commentary to break explanations and surveillance triage, with tools granted per role and a human sign-off before anything reaches a client, a counterparty or the order management system. Each step is logged with the requesting user, the model, the retrieved sources and the tool calls made.

The VDF AI Router registers local open-weight models running on vLLM, Ollama or another OpenAI-compatible endpoint, and can admit an approved cloud model for public-information tasks such as filing summaries, applying routing rules, budgets and per-call audit across both. On-premises sign-in uses Microsoft Entra ID natively, while Okta, Keycloak and other SAML or OIDC identity providers connect through an SSO-aware reverse proxy, and role-based access control is included on every plan. A manager can start with one team and one workflow under a three-month on-premises platform trial, then size seat bundles from what that team learns.

The finance and banking solution, the hub of finance AI use cases and the finance executive brief describe the wider financial services architecture. The free AI readiness assessment shows whether data access, governance or skills will hold back a first pilot.

This guide is general information and not legal, regulatory or investment advice.