2026 Outlook Enterprise Investment Management AI Transformation

From productivity pilots
to fiduciary-grade
transformation.

Three years after ChatGPT, investment management has reached AI ubiquity in pilots but not AI value in decisions. The differentiator between adopters and winners is workflow and governance redesign — not tool count, and not headcount cuts. Most firms have scaled productivity tools while leaving investment governance untouched.

$147T

global AUM at mid-2025
(McKinsey, 2025)

47%

of top-500 managers report active AI investment
(WTW Thinking Ahead Inst., 2025)

78%

allocate <10% of tech budget to AI
(WTW Thinking Ahead Inst., 2025)

11%

of agentic AI workloads in production
(Deloitte, 2026 IM Outlook)

10 reference tables · expandable below

Every analytical table from the underlying research report — value chain, research workflow, portfolio & trading, operating model, roles, model risk, legal, red lines, cases, governance — is available below as a click-to-expand data view. Closed by default to keep the briefing scannable.

01Value Chain 02Research Workflow 03Portfolio & Trading 04Operating Model 05Roles 06Model Risk & Bias 07Legal 08Red Lines 09Cases 10Governance

01 The Thesis

Three positions this report defends.

Investment firms treating AI as a productivity overlay capture some cost benefit but no durable advantage. The firms pulling ahead are redesigning the investment workflow and embedding fiduciary-grade governance — and the regulatory floor is rising faster than most compliance teams have provisioned for.

Productivity ROI is real but commoditizing.

Morgan Stanley's reported 98% advisor adoption and JPMorgan's cited AI value (roughly $1.5–2.0B annually in its investor-day materials) are credible directionally — but within 24 months these become table stakes on the same vendor stacks. The lasting advantage is what firms build into the investment process, not the cost they remove. Treat all such figures as company commentary, not audited fact.

ii.

The value pool is in decision-quality, not speed.

Reaching Level 4–5 maturity requires the CIO to spend capital on research-workflow redesign, IC evidence packs, and fiduciary-grade model governance — not on tool selection. Almost every published AI metric is firm-side productivity; client-outcome evidence is nearly absent.

iii.

AI policies built in 2023 are already out of date.

SEC AI-washing enforcement (Delphia / Global Predictions), FINRA RN 24-09, ESMA's MiFID II statement, and the EU AI Act have redrawn the red lines. The SEC's withdrawal of the predictive-data-analytics proposal is not deregulation — existing fiduciary, conduct, and Marketing-Rule obligations still apply.

02 Adoption vs. Value

Nearly half are investing in AI. Few are winning with it.

The instructive gap is between adoption and value capture. Most firms surveyed sit at functional pilots; very few have redesigned the investment workflow or made any change to investment governance. Workflow and governance redesign is the variable that separates the cohorts.

The AI adoption gap in investment management

McKinsey · Deloitte · WTW Thinking Ahead Inst.

Top-500 managers reporting active AI investmentWTW Thinking Ahead Institute, 2025

47%

Firms expecting GenAI ROI within one yearDeloitte enterprise AI survey

38%

Firms seeing significant GenAI ROI todayDeloitte, 2025

15%

Agentic AI workloads in productionDeloitte, 2026 IM Outlook

11%

Tech budget allocated to AI — most managers

<10%

78% of the top-500 asset managers allocate less than 10% of their technology budget to AI, even as 47% report active AI investment. Adoption intent runs well ahead of committed spend.

Productivity lift on high-impact AI use cases

25–40%

McKinsey's primary 2025 report puts productivity lifts at roughly 25–40% on high-impact use cases (e.g., PM copilots, code copilots), with mid-single-digit gains in investment research specifically. A widely repeated "~8% across the value chain" figure traces to secondary commentary, not McKinsey's report.

03 Workflow Redesign

Four levels of AI in investment work.

The four are not interchangeable. Most firms stall at level two — adding AI to existing workflows — and never restructure the investment process itself, the investment committee, or the organization around it.

Task automation

Faster, cheaper, same workflow. AI summarizes earnings calls and filings; analyst still reviews.

Workflow augmentation

Human + AI in the same workflow. Advisor copilot drafts; advisor approves. Morgan Stanley AI@MS Assistant.

III

Workflow reinvention

The investment process itself changes. AI-assisted thesis generation with a structured devil's-advocate challenge step before the IC.

Operating-model transformation

The organization changes around the new workflow — new roles (investment AI product owner, knowledge-graph manager, AI model-risk partner), IC evidence packs, and fiduciary-grade governance. Schroders Capital's GAiiA and IC challenger agent are an explicit example in private markets.

04 Enterprise Cases

What leaders have actually deployed.

A balanced read pairs success cases with cautionary ones. Figures below are company- or vendor-reported and are not independently audited. The full 12-case comparison is in reference table 09.

Morgan Stanley

Wealth

AI @ Morgan Stanley Assistant and Debrief, built on OpenAI, giving advisors retrieval over a large internal research corpus.

~98%

reported advisor adoption of the Assistant

20–80%

reported shift in advisor document access

Company-reported; not independently audited. A Level-3 productivity deployment, not a change to investment governance.

JPMorgan

Bank / AM

LLM Suite rolled out enterprise-wide as a general-purpose assistant across research, operations, and support functions.

~200K

LLM Suite users (JPMorgan investor-day materials)

$1.5–2.0B

annual AI value, per investor-day presentation

200K-user figure is well-supported by JPMorgan itself; the value range is company commentary (the narrower "$1.5B" and "30–40% efficiency" figures appear mostly in secondary coverage) and is not independently audited.

Schroders Capital

Private markets

GAiiA GenAI investment advisor plus an explicit IC challenger agent used across 40+ deal evaluations.

40+

deal evaluations supported

workflow-redesign maturity in private markets

CIO Nils Rode on record: every investment decision is still made solely by investment professionals.

Arup (cautionary)

Deepfake fraud

A finance employee was deceived by a real-time AI-impersonated video conference of the CFO and colleagues, authorizing a series of transfers.

~$25.6M

lost (≈HK$200M), January 2024

Live

real-time deepfake of multiple known colleagues

FT / CNN / Fortune reporting; Arup CIO Rob Greig publicly confirmed. Illustrates deepfake risk to executive approval workflows.

05 Maturity Model

Five levels — and where firms actually sit.

Progression is not linear and is not guaranteed by spending. Many firms have moved to Level 3 in productivity tools while leaving investment governance at Level 1–2. Genuine transformation requires advancement on capability, governance, data, and accountability together.

Level 01

Ad hoc experimentation

Individuals use public or sanctioned tools opportunistically. No inventory, no approval workflow, no measurement. Risk from MNPI in prompts, hallucination, and shadow IT.

Level 02

Functional pilots

Named pilots with sponsors and metrics in research, reporting, or surveillance. Siloed; productivity claims anecdotal. Most surveyed firms sit here in 2025–26.

Level 03

iii

Scaled use cases

Selected use cases deployed at scale with approved tools, citation requirements, and supervisory review. ROI reported as productivity, not decision quality. Morgan Stanley, JPMorgan.

Level 04

Workflow redesign

AI embedded in the investment workflow. Research redesigned around AI-assisted theses with a challenge step; IC evidence packs; manager-DD knowledge graph. Schroders Capital.

Level 05

Decision-intelligent platform

Every material recommendation has a traceable evidence pack, documented challenge, recorded decision, and feedback loop. Fiduciary review is auditable. No firm publicly demonstrates this end-to-end.

Where most firms sit: Level 3 on productivity, Level 1–2 on investment governance. The hard, valuable work — Levels 4 and 5 — requires redesigning the research process, the investment committee, the data layer, and accountability at once. Spending alone does not move a firm up the model.

06 AI Red Lines

Where AI should be prohibited or tightly restricted.

Four tiers, from outright prohibition to enhanced-control. These are the workflows where regulators have penalized firms and where reputational and fiduciary damage concentrates. The full 21-item table is reference table 08.

01Prohibited

Autonomous decisions

Autonomous portfolio recommendation engines with no human sign-off
Autonomous trading agents acting without an algorithm inventory and kill-switch
MNPI or client PII pasted into unmanaged public LLMs

02High restriction

Client-facing AI

AI-generated client communications bypassing Marketing-Rule review
AI risk-profile classification without explicit client input
Fabricated/synthetic testimonials or hypothetical performance (genuine testimonials allowed only under the Marketing Rule's disclosure & oversight conditions)

03Enhanced control

Research & advice

Publishing AI-generated research without analyst certification
Next-best-action engines optimizing for firm over client
AI-drafted RFP/DDQ responses without substantiation

04Monitor

Operational

AI exception triage that auto-corrects client-impacting items
Vendor/model concentration without contingency
Unlogged AI use in supervised communications

07 Regulatory Floor

The regulatory floor is rising.

The posture across jurisdictions is technology-neutral: existing fiduciary, conduct, and marketing rules already apply to AI. The full legal matrix is reference table 07.

2023Jul

SEC

Predictive Data Analytics proposal

Proposed eliminating or neutralizing conflicts in investor-interaction technology — next-best-action, gamification, behavioral nudges. (Rel. 34-97990.)

2024Mar

SEC Enforcement

Delphia & Global Predictions — AI-washing

$225K and $175K penalties for false/misleading AI claims, under Advisers Act §206, Marketing Rule 206(4)-1, and Compliance Rule 206(4)-7. (Press release 2024-36.)

2024May

ESMA

Statement on AI in retail investment services

MiFID II conduct, suitability, organisational, and best-interest obligations apply to AI use with retail clients.

2024Jun

FINRA

Regulatory Notice 24-09

Technology-neutral: all FINRA rules — supervision, communications, books and records, cyber — apply to AI and GenAI/LLMs.

2024Nov

FSB

Financial Stability Implications of AI

Four vulnerabilities: third-party concentration, market correlation, cyber, and model/data risk.

2025Jun

SEC

PDA proposal withdrawn (33-11377)

Not deregulation. Underlying concerns endure under existing fiduciary, conduct, and Marketing-Rule obligations.

08 12–24 Month Plan

A sequenced CIO action agenda.

Controls before scaling; investment-decision use cases after productivity use cases. Smaller firms can compress the phases but should not invert them.

Months 0–3

Foundations

Investment-specific AUPPublish an AI acceptable-use policy distinct from enterprise IT, covering MNPI, holdings, client data, and models.

Governance committee + inventoryStand up a CIO-chaired AI governance committee; inventory and risk-tier every use case.

Approved tools + baselineDefine an approved-tools list, block unmanaged LLMs, and measure a productivity baseline so ROI is falsifiable.

Months 3–9

Scaled productivity

Copilot with controlsDeploy an advisor/PM copilot on a vetted RAG corpus with citation requirements, a compliance queue, and supervision logging.

Summarization + RFP/DDQRoll out earnings-call and filing summarization with mandatory analyst review; an RFP/DDQ assistant with Marketing-Rule pre-clearance.

Model inventory + testingBuild a model inventory; run hallucination, bias, and drift testing with thresholds and remediation triggers.

Months 9–18

Workflow redesign

Research + IC redesignAI handles sourcing and synthesis; analysts handle judgment. Require a structured devil's-advocate challenge on every thesis; redesign the IC evidence pack.

Manager-DD knowledge graphLink funds, sponsors, deals, and flags; AI retrieves and synthesizes, humans conclude.

Reskill, don't just enableTrain analysts and PMs on prompting, source verification, and structured challenge.

Months 18–24

Operating model

Reshape rolesFewer junior task workers; new roles — investment AI product owner, data product manager, knowledge-graph manager, AI model-risk partner.

Decision-intelligence layerRecord thesis, evidence, challenge, decision, decider, and outcome for every material decision; build the feedback loop.

Recertify + board reviewRecertify production use cases annually; brief the board on maturity, controls, disclosures, and vendor concentration.

10 Reference Tables · Expandable

The full evidence, on demand.

Every analytical table from the underlying research is below as a click-to-expand data view. The Roles & skills table is open by default — it's the most relevant evidence for individual contributors on how AI augments specific investment roles. The other nine are closed to keep the briefing scannable; open one, several, or all at once.

Twenty-four use cases mapped by function, AI type, maturity, value, and risk. The most mature are document summarization and predictive analytics; the most consequential shifts are in research-thesis challenge, IC decision support, and manager due diligence.

Function	Use case	AI type	Maturity	Business value	Key risk	Example / source
Investment research	Earnings-call & filing summarization	Generative (LLM)	Scaled	Analyst time saved; coverage breadth	Hallucinated facts; missed nuance	Morgan Stanley Debrief; broad adoption
Investment research	Thematic screening & idea generation	Generative + retrieval	Scaling	Wider opportunity set	Crowding; spurious patterns	Multiple managers
Investment research	Devil's-advocate thesis challenge	Generative (agentic)	Emerging	Decision quality; bias reduction	Over-reliance on AI challenge	Schroders Capital IC challenger
Quant research	ML factor & signal discovery	Predictive ML	Scaled	Risk-premia prediction	Backtest overfitting; regime shift	Gu, Kelly & Xiu (RFS 2020); Two Sigma
Portfolio construction	Optimization & scenario analysis	Predictive + optimization	Scaling	Allocation efficiency	Causal overclaims; opacity	BlackRock Aladdin
Portfolio construction	Tax-aware / direct indexing	Optimization	Scaling	After-tax alpha (simulated)	Wash-sale, suitability, tracking error	Vanguard, Schwab Personalized Indexing
Trading	Execution analytics & TCA	Predictive + reinforcement	Scaled	Lower implementation shortfall	Model-driven correlation in stress	Goldman Sachs; State Street
Trading	Algo selection & smart routing	Reinforcement	Scaling	Execution quality	Autonomous action without controls	Buy-side desks
Operations	Reconciliation & exception triage	Predictive + LLM	Scaling	Cycle time; fewer breaks	Auto-correction of client-impacting items	State Street Alpha (vendor-stated)
Operations	Corporate-actions processing	LLM + rules	Emerging	STP rates	Misclassification	Custodian platforms
Risk management	Anomaly & outlier detection	Predictive	Scaling	Faster detection	False positives; opacity	State Street (~25x, vendor-stated)
Risk management	Scenario & stress narrative generation	Generative	Emerging	Faster risk reporting	Hallucinated assumptions	Multiple managers
Advisor support	Advisor copilot / knowledge retrieval	Generative + retrieval	Scaled	Meeting prep; doc access	Best-interest / Reg BI exposure	Morgan Stanley AI@MS Assistant
Advisor support	Meeting notes & follow-up drafting	Generative	Scaled	Advisor time saved	Recordkeeping; supervision	UBS Red; M365 Copilot
Client reporting	Commentary & performance narratives	Generative	Scaling	Reporting cycle time	Marketing Rule; GIPS	Multiple managers
Distribution	RFP / DDQ response drafting	Generative + retrieval	Scaling	RFP cycle time	Unsubstantiated claims	Buy-side sales desks
Manager selection	Manager DD knowledge graph	Retrieval + graph	Emerging	DD breadth & recall	Stale or wrong linkage	Allocators, fund-of-funds
Private markets	Deal screening & memo drafting	Generative (agentic)	Emerging	Deal throughput	Overclaim; thin diligence	Schroders Capital GAiiA
ESG / stewardship	Disclosure extraction & scoring	LLM + predictive	Scaling	Coverage; consistency	Greenwashing; data quality	Multiple managers
Compliance	Communications surveillance	Predictive + LLM	Scaled	Detection; coverage	Model bias; missed intent	Surveillance vendors
Compliance	Pre-publication marketing review	LLM + rules	Emerging	Review cycle time	False sense of automated compliance	Saifr and peers
Client service	Service triage & FAQ assistants	Generative	Scaling	Response time	Wrong account-specific answers	Wealth platforms
Product	Personalization & next-best-action	Predictive + agentic	Emerging	Engagement; retention	Firm-vs-client conflict (PDA concerns)	Wealth platforms
Enterprise	Foundation-model copilots (M365 etc.)	Generative	Scaled	Cross-function productivity	Vendor / model concentration	UBS (~50K Copilot seats by Q1 2025)

Eleven research and decision workflows mapped from current-state pain through AI-enabled redesign, with human/AI roles, new controls, and outcome metrics. The discipline is that AI handles sourcing and synthesis while analysts retain judgment and conclusions.

Workflow	Current-state pain	AI-enabled redesign	Human role	AI role	New controls	Metrics
Idea sourcing	Limited coverage; analyst bandwidth	Continuous screening of filings, transcripts, news, alt-data	Analyst sets thesis & judgment	Surface candidates, summarize	Source-citation requirement; provenance log	Coverage breadth; ideas per analyst
Document analysis	Hours per filing / transcript	LLM extraction & summarization with links to source	Analyst verifies & interprets	Extract, summarize, tag	Mandatory analyst review; no unsourced claims	Time per document; error rate
Thesis construction	Confirmation bias; uneven rigor	AI-assisted thesis with explicit assumptions	Analyst owns the thesis	Structure, surface counter-evidence	Assumptions logged & testable	Thesis completeness; revision rate
Devil's-advocate challenge	Weak or absent challenge step	Challenger agent generates the bear/bull counter-case	Analyst responds on record	Generate structured challenge	Challenge attached to every recommendation	% theses revised post-challenge
Financial modeling	Manual, error-prone	AI-assisted model scaffolding & checks	Analyst owns assumptions	Draft, sanity-check, flag	Model-review sign-off	Model build time; check exceptions
Quant signal research	Overfitting risk	ML signal discovery with OOS discipline	Researcher designs & validates	Search, test, rank	Out-of-sample & regime testing	OOS performance; decay rate
IC evidence pack	Inconsistent, narrative-driven	Standardized AI-prepared pack with citations + challenger note	PM / IC decide	Assemble evidence & challenge	Traceability of every claim	IC cycle time; evidence traceability
Manager due diligence	Siloed documents; weak recall	Knowledge graph across funds, sponsors, deals, flags	DD team concludes	Retrieve, link, synthesize	Linkage validation; stale-data alerts	DD cycle time; flag recall
ESG / stewardship	Manual disclosure review	Extraction & scoring with human override	Stewardship lead judges	Extract, score, flag	Greenwashing check; data lineage	Coverage; override rate
Macro / strategy	Fragmented inputs	AI synthesis of macro data & scenarios	Strategist owns the call	Synthesize, simulate	Assumption transparency	Forecast tracking; scenario coverage
Research publication	Slow editing; compliance lag	AI drafting with analyst certification	Analyst certifies content	Draft, format, check	Analyst certification; Marketing-Rule review	Time to publish; compliance incidents

Eleven portfolio-construction, trading, and operations workflows. The recurring control is a human-approval gate on any client-impacting action and an algorithm inventory for anything that touches execution.

Workflow	AI-enabled redesign	Human role	AI role	New controls	Metrics
Asset allocation	AI scenario modeling & optimization	CIO / IC owns allocation	Simulate, optimize, stress	Assumption transparency; override log	Allocation efficiency; tracking error
Security selection	ML-ranked candidate lists	PM decides	Rank, explain, flag	Explainability; human sign-off	Hit rate; turnover
Tax-aware rebalancing	Direct indexing with tax-lot logic	Advisor / PM approves first trade	Optimize lots, harvest losses	Wash-sale, IPS, suitability guardrails	After-tax alpha (simulated); tracking error
Risk budgeting	AI-assisted factor & risk decomposition	Risk officer judges	Decompose, attribute	Model validation; limits	Risk-limit breaches; attribution accuracy
Execution	Predictive TCA & smart routing	Trader oversees	Predict cost, route	Algorithm inventory; kill-switch	Implementation shortfall
Algo selection	Reinforcement-learned routing	Desk head approves strategies	Select, adapt	Pre-trade limits; monitoring	Execution quality vs benchmark
Rebalancing ops	Automated drift detection & proposals	PM approves trades	Detect drift, propose	Approval gate; audit trail	Drift; rebalancing latency
Reconciliation	AI exception triage	Ops approves corrections	Match, flag, propose	No auto-correct on client items	Break rate; resolution time
Corporate actions	LLM extraction + rules	Ops validates	Classify, populate	Dual control on high-value events	STP rate; error rate
Collateral / liquidity	Predictive forecasting	Treasury decides	Forecast, alert	Limit framework	Forecast accuracy; buffer breaches
Performance & attribution	AI-generated attribution narratives	Performance officer reviews	Compute, narrate	GIPS-conformant policies	Reporting cycle time; restatements

Seventeen dimensions of the investment operating model, contrasting the traditional state with the AI-enabled state and the governance shift each requires.

Dimension	Traditional state	AI-enabled state	Governance shift	Owner
Research process	Manual sourcing & synthesis	AI sourcing/synthesis; analyst judgment	Source citation; analyst certification	Heads of Research
Analyst role	Extraction-heavy	Judgment-heavy	Reskilling; new career path	CIO / HR
Investment committee	Narrative memos	Standardized evidence packs + challenge	Traceability of recommendations	CIO / IC
Idea pipeline	Narrow, bandwidth-bound	Broad, continuously screened	Crowding & quality monitoring	Heads of Research
Portfolio construction	Spreadsheet-driven	AI scenario & optimization support	Explainability; override logs	PMs / CIO
Trading desk	Manual algo selection	AI execution analytics	Algorithm inventory; kill-switch	Head of Trading
Operations	Processing-centric	Exception-management-centric	Approval gates on client items	COO
Risk management	Periodic reporting	Continuous AI monitoring	Model validation; SR 11-7-style	CRO
Compliance	End-of-process review	Embedded pre-clearance gates	Automated Marketing-Rule checks	CCO
Advisor support	Manual prep & research	Copilot with RAG corpus	Supervision logging; Reg BI	Head of Wealth
Client reporting	Manual drafting	AI-generated commentary	GIPS; Marketing-Rule review	Performance / Marketing
Manager DD	Document silos	Knowledge graph	Linkage validation	Manager Research
Data layer	Fragmented	Curated investment data products	Lineage & quality controls	CDO
Technology	Legacy systems	Enterprise AI platform + RAG	Vendor concentration management	CTO
Vendor strategy	Point solutions	Platform + model portfolio	Concentration & contingency	CTO / Procurement
Decision record	Sparse	Thesis-evidence-challenge-decision log	Auditable fiduciary trail	CIO / CCO
Talent model	Pyramid of juniors	Fewer juniors; specialist roles	Judgment-development design	CIO / HR

Eleven investment roles mapped from today's profile to the AI-augmented profile, with the tasks AI absorbs, the judgment that stays human, the new skills required, and the open question for each role.

Role	AI absorbs	Stays human	New skills	Net direction	Open question
Research analyst	Extraction, summarization, first-draft models	Thesis, judgment, conviction	Prompting, source verification, structured challenge	Fewer juniors; higher judgment bar	Where does junior judgment develop?
Portfolio manager	Scenario modeling, candidate ranking	Allocation decisions, conviction sizing	Model literacy, override discipline	Augmented, not replaced	Does AI commoditize edge?
Quant researcher	Signal search, backtesting scaffolds	Hypothesis design, OOS validation	LLM tooling, causal discipline	Higher leverage	Crowding into shared signals?
Trader	TCA, routing, monitoring	Strategy choice, market judgment	Algo governance, exception handling	Oversight-centric	Correlation risk under stress?
Investment-operations analyst	Matching, reconciliation, triage	Exception resolution, client-impact calls	Exception management, AI QA	Processing → exceptions	Auto-correction limits?
Financial advisor	Meeting prep, notes, retrieval	Relationship, suitability judgment	Copilot fluency, supervision awareness	Capacity expansion	Best-interest accountability?
Manager-research analyst	Document linkage, recall	Diligence conclusions	Knowledge-graph use	Broader coverage	Stale-linkage risk?
Risk officer	Anomaly detection, reporting	Risk-appetite judgment	Model risk management	Continuous oversight	How to validate opaque models?
Compliance officer	Surveillance, pre-clearance checks	Conduct judgment, escalation	AI governance, model inventory	Embedded upstream	Automated-compliance false comfort?
Client-reporting specialist	Commentary drafting, formatting	Accuracy & messaging judgment	Marketing-Rule literacy	Faster cycles	GIPS & disclosure controls?
New: Investment AI product owner	—	Workflow design, control ownership	Product, governance, investment domain	Net-new role	Where does it report?

Twelve model-risk and bias failure modes specific to investment management, with where they bite, who owns them, and the control that contains them. Anchored to NIST AI RMF and SR 11-7-style model governance.

Risk	Where it bites	Mechanism	Consequence	Control	Owner	Reference
Backtest overfitting	Quant research	Mining noise as signal	OOS underperformance	OOS & regime testing; holdout discipline	Head of Quant	Lopez de Prado; Gu/Kelly/Xiu
Regime shift	Models trained on one regime	Distribution change	Sudden drawdown	Regime monitoring; circuit breakers	CRO / PM	Renaissance ~20% loss 2020
Hallucination	Research, reporting	Fabricated facts/cites	Wrong decisions; misstatement	Source-citation; analyst review	Heads of Research	NIST AI 600-1
Crowding / correlation	Shared models & data	Homogeneous behavior	Systemic stress amplification	Model diversity; stress testing	CRO	FSB Nov 2024
Proxy discrimination	Client segmentation	Proxies for protected class	Unfair treatment; legal	Bias audit; fair-treatment checks	CCO	Bartlett et al. (JFE 2022)
Data leakage	Signal research	Look-ahead in features	Illusory performance	Point-in-time data; lineage	CDO / Quant	Standard ML practice
Model drift	Production models	Silent degradation	Decaying quality	Drift thresholds; recertification	Model-risk partner	SR 11-7
Opacity	Risk & selection	Unexplainable outputs	Cannot justify to fiduciary standard	Explainability; documented basis	CRO / CIO	SR 11-7; Advisers Act §206
Automation bias	Analysts, advisors	Over-trust in AI	Unchallenged errors	Mandatory challenge; training	CIO	Human-factors literature
Vendor / model concentration	Enterprise	Few base models	Single point of failure	Portfolio of models; contingency	CTO	FSB Nov 2024
Conflicts in client tools	Next-best-action	Optimizing firm interest	Fiduciary breach	Conflict elimination/neutralization	CCO / GC	SEC PDA concerns
Deepfake / social engineering	Approvals, payments	AI-impersonated audio/video	Fraudulent transfers	Call-back; multi-party verification	CISO / COO	Arup (FT/CNN 2024)

Fourteen regulations and standards mapped to their investment-management implication, the high-risk use cases, required controls, and owner. The posture is technology-neutral: existing rules already apply to AI.

Regulation / standard	Implication	High-risk use cases	Required controls	Owner	Source
Advisers Act §206 fiduciary duty	Duty of care + loyalty applies to AI-influenced advice	AI advice without review; NBA optimizing for firm	Documented basis; conflicts addressed	CCO / GC	Advisers Act; SEC IA-5248
Reg Best Interest	Best-interest standard covers AI-generated recommendations	Advisor copilot recommendations without analysis	Conflict ID; mitigation; disclosure	CCO / Supervision	SEC Reg BI
SEC Marketing Rule 206(4)-1	Applies to AI-generated commentary & advertising	AI client commentary; AI RFP responses	Pre-publication review; substantiation	CCO / Marketing	Delphia / Global Predictions 2024
FINRA RN 24-09	All FINRA rules apply to AI/GenAI/LLMs	Communications, supervision, MNPI, cyber	Apply existing supervision & recordkeeping	CCO / CTO	FINRA RN 24-09 (Jun 2024)
SEC PDA proposal (withdrawn)	Withdrawn but concerns endure under existing rules	NBA engines; gamification; nudges	Conflict elimination/neutralization	CCO / GC	SEC 33-11377 (Jun 2025)
FINRA suitability + Reg BI	AI recommendations must remain suitable	AI-classified risk tolerance	Documented suitability; client input	CCO / Supervision	FINRA Rules; Reg BI
GIPS Standards	AI performance commentary must conform	AI commentary implying composite performance	GIPS-conformant policies	Performance Officer	CFA Institute GIPS
CFA Code & Standards	I(C), V(A), V(B), III(C) implicated	Uncertified AI research; no diligence	Analyst certification; documented diligence	CIO / Research	CFA Institute (eff. Jan 2024)
EU AI Act (2024/1689)	Risk-tiered; some financial use cases high-risk	Creditworthiness; certain client-facing AI	High-risk obligations; documentation	GC / CCO (EU)	Reg (EU) 2024/1689
ESMA AI statement	MiFID II conduct & best-interest apply	AI in retail investment services	Suitability; organisational controls	CCO (EU)	ESMA (May 2024)
NIST AI RMF + 600-1	Governance scaffold for model risk	All production AI	Map-measure-manage; GenAI profile	CRO / CTO	NIST (2023 / 2024)
SR 11-7 (by analogy)	Model-risk lifecycle standard	Models influencing decisions	Validation; inventory; recertification	Model-risk partner	Fed / OCC (2011)
FCA Consumer Duty	Outcomes-focused conduct (UK)	AI-driven retail investment services	Demonstrate good outcomes	CCO (UK)	FCA PS22/9
NYDFS Part 500	Cybersecurity requirements	AI systems handling client data	Cyber controls; incident reporting	CISO	23 NYCRR 500

Fifteen use cases by tier — prohibited, high-restriction, enhanced-control — with the rationale, the minimum control if permitted at all, and the owner. These are where regulators have penalized firms and where fiduciary damage concentrates.

Use case	Tier	Rationale	Minimum control (if permitted)	Owner	Reference
MNPI / client PII in public LLMs	Prohibited	Confidentiality, MNPI, privacy breach	Block; enterprise tools only	CISO / CCO	FINRA RN 24-09
Autonomous portfolio recommendations	Prohibited	Fiduciary accountability	Human sign-off required	CIO / CCO	Advisers Act §206
Autonomous trading agents	Prohibited	Market & correlation risk	Algorithm inventory + kill-switch	Head of Trading	FSB Nov 2024
Fabricated / synthetic testimonials	Prohibited	False/misleading communications. (Genuine testimonials are permitted under the Marketing Rule's disclosure, oversight & disqualification conditions — AI assistance per se is not banned.)	No fabrication; genuine testimonials only with disclosures + oversight	CCO / Marketing	Rule 206(4)-1
AI risk-profiling without client input	High restriction	Suitability; best interest	Explicit client input required	CCO / Supervision	Reg BI; FINRA
Client comms bypassing review	High restriction	Marketing Rule; supervision	Pre-publication review queue	CCO / Marketing	Rule 206(4)-1
Next-best-action for firm benefit	High restriction	Firm-vs-client conflict	Conflict elimination/neutralization	CCO / GC	SEC PDA concerns
Hypothetical performance via AI	High restriction	Marketing Rule conditions	Strict Marketing-Rule controls	CCO	Rule 206(4)-1
Uncertified AI research publication	Enhanced control	Diligence; misrepresentation	Analyst certification	Heads of Research	CFA Code; Advisers Act
AI-drafted RFP/DDQ without check	Enhanced control	Unsubstantiated claims	Substantiation; version control	Distribution / CCO	Rule 206(4)-1
Auto-correcting client-impacting ops	Enhanced control	Error propagation	Human approval gate	COO	Internal controls
Opaque models in selection/risk	Enhanced control	Cannot justify to fiduciary standard	Explainability; documented basis	CRO / CIO	SR 11-7
Single-vendor model dependence	Enhanced control	Concentration risk	Model portfolio; contingency	CTO	FSB Nov 2024
Unlogged AI in supervised comms	Enhanced control	Recordkeeping; supervision	Logging; surveillance coverage	CCO / Supervision	FINRA RN 24-09
Executive payment approvals over video	Enhanced control	Deepfake fraud	Call-back; multi-party verification	CISO / COO	Arup (FT/CNN 2024)

Twelve enterprise cases — ten deployments and two cautionary — with the application, reported metric, maturity level, and the critical caveat. All metrics are company- or vendor-reported and not independently audited.

Firm	Segment	Application	Reported metric	Maturity	Stance	Caveat
Morgan Stanley	Wealth	AI@MS Assistant & Debrief (OpenAI)	~98% advisor adoption; 20→80% doc access	Level 3	Positive	Company-reported; productivity, not governance
JPMorgan	Bank / AM	LLM Suite enterprise-wide	~200K users (well-supported); $1.5–2.0B annual AI value (investor-day)	Level 3	Positive	200K solid; value range is company commentary, not audited
BlackRock	Asset mgmt	Aladdin Copilot, eFront Copilot, Asimov	~$2T AUM growth attributed (CEO)	Level 3–4	Positive	CEO attribution; not isolable to AI
UBS	Wealth / bank	M365 Copilot + proprietary "Red"	~50K Copilot; ~30K Red (by Q1 2025)	Level 3	Positive	Vendor-supported customer story
Schroders Capital	Private markets	GAiiA + IC challenger agent	40+ deal evaluations	Level 4	Positive	Decisions still made solely by professionals
Vanguard	Wealth / AM	Personalized indexing	Marketing: "up to 1–2%" after-tax alpha; Vanguard Research: 0.47–1.27% TLH alpha	Level 3	Positive	"Up to 1–2%" is a marketing ceiling; research figure is lower and must be scaled by taxable-equity share
Schwab	Wealth	Personalized indexing	Lowered minimums; broad access	Level 3	Positive	Democratization claim
Two Sigma	Quant	Deep ML + LLM research/execution	Not disclosed	Level 4	Neutral	Limited primary sourcing
State Street	Asset servicing	State Street Alpha; anomaly detection	100% of true exceptions caught; ~87% fewer flagged items = 25x productivity gain	Level 3	Positive	Vendor-stated; single use case (market-value data validation)
Goldman Sachs	Bank / AM	AI in risk & execution	Not quantified publicly	Level 3	Neutral	Limited disclosure
Delphia / Global Predictions	RIA	AI-washing in marketing	$225K + $175K penalties	—	Cautionary	SEC enforcement (2024-36)
Arup	Cross-industry	Deepfake video-conference fraud	~$25.6M lost (Jan 2024)	—	Cautionary	FT/CNN/Fortune; CIO confirmed

Seventeen governance controls across the AI lifecycle, with the control objective, the artifact that evidences it, the owner, and the cadence. Together these form a fiduciary-grade control model anchored to NIST AI RMF and SR 11-7-style model governance.

Control area	Control objective	Evidence artifact	Owner	Cadence	Reference
Policy	Investment-specific AI acceptable-use policy	Published AUP	CIO / CCO	Annual review	Internal
Governance body	CIO-chaired AI committee	Charter; minutes	CIO	Monthly	Internal
Use-case inventory	All use cases catalogued & risk-tiered	Inventory register	AI product owner	Quarterly	NIST AI RMF
Approved tools	Sanctioned tool list; public LLMs blocked	Tool register; controls	CTO / CISO	Quarterly	FINRA RN 24-09
Model inventory	Every decision-influencing model logged	Model register	Model-risk partner	Continuous	SR 11-7
Model validation	Independent validation before production	Validation report	CRO	Pre-deployment	SR 11-7
Bias & fairness	Testing on protected proxies	Bias-audit results	CCO	Quarterly	NIST; Bartlett et al.
Hallucination testing	Thresholds & remediation triggers	Test logs	AI product owner	Continuous	NIST AI 600-1
Drift monitoring	Detect silent degradation	Drift dashboards	Model-risk partner	Continuous	SR 11-7
Source citation	No unsourced AI claims in research	Citation logs	Heads of Research	Per output	CFA Code
Analyst certification	Human certifies published research	Certification record	Analysts	Per publication	CFA Code
Marketing-Rule review	Pre-publication review of client content	Review queue logs	CCO / Marketing	Per item	Rule 206(4)-1
Supervision logging	Advisor copilot output supervised	Supervision dashboards	Supervision	Continuous	FINRA; Reg BI
Decision record	Thesis-evidence-challenge-decision log	Decision-intelligence log	CIO / IC	Per decision	Advisers Act §206
Vendor concentration	Model portfolio & contingency	Concentration analysis	CTO	Semi-annual	FSB Nov 2024
Recertification	Annual sign-off on production use cases	Recertification record	CIO / CCO / CRO	Annual	SR 11-7
Board reporting	Posture, controls, disclosures, concentration	Board pack	CIO	Annual	Internal

09 · So What

What CIOs should do differently now.

Stop benchmarking your AI adoption by tool count. Start benchmarking by the question every regulator, professional body, and consulting firm in the bibliography is converging on: how much of your investment workflow has been redesigned around AI, and is every material recommendation supported by a traceable, challenged, fiduciary-grade decision record?

By 2027, the difference between investment functions that thrived through the AI transition and those that did not will not be access to models — every CIO will have access to the same frontier models, the same platforms, and many of the same vendors.

The difference will be (1) how aggressively the research workflow, investment committee, and operating model were redesigned; (2) how rigorously fiduciary duty, conduct, model risk, and conflicts were embedded upstream rather than bolted on downstream; and (3) whether the CIO treated AI productivity figures as company commentary to be verified, not audited fact to be repeated.

The CIOs who will lead enterprise AI transformation are the ones who treat AI not as a productivity story to tell the board, but as the most important investment-governance decision their firm will make this decade.

End of briefing · The CIO's AI Decision · v1

Your future is our focus.™

From productivity pilotsto fiduciary-gradetransformation.