THE AI TAXONOMY

LO 1: CURRENT AND FUTURE USES

The IMF reading draws a deliberate and exam-relevant distinction between two tiers of AI. Candidates must understand what separates them and how each is currently used.

AI / ML MODELS (ESTABLISHED)

Well-established predictive analytics including shallow neural networks, clustering algorithms, decision trees, NLP, and textual analysis tools. Used in financial services for ~20 years.

• Signal generation and return forecasting
• Anti-money laundering and fraud detection
• RegTech compliance automation
• Robo-advising and portfolio rebalancing

SOPHISTICATED AI MODELS (ADVANCED)

More recent and complex architectures including deep neural networks, reinforcement learning agents, and LLMs. This tier includes Generative AI (GenAI).

• Real-time parsing of central bank communications
• Autonomous signal generation from alt data
• Code generation and strategy prototyping
• Synthetic data for illiquid market modeling

Key exam point: Most current adoption is still in the first tier. Sophisticated AI use remains nascent and is primarily a medium- to long-term concern.

CURRENT AI USE CASES ACROSS THE INVESTMENT PROCESS

LO 1: CURRENT AND FUTURE USES

AI adoption is uneven across the investment workflow. It is most integrated in early-stage analytical tasks and least used in final execution and asset allocation decisions.

CLIENT PROFILING & ASSET ALLOCATION

AI analyzes unstructured client data to model risk preferences and simulate portfolio outcomes. Deep learning supports dynamic multiperiod portfolio optimization and clustering of asset correlations.

SECURITY SELECTION & SIGNAL GENERATION

NLP and sentiment analysis extract signals from regulatory filings, social media, and news. Feature extraction identifies relative value and price dislocations. Over 43% of surveyed managers report alternative data as already integrated.

TRADE EXECUTION

Structured execution algorithms minimize market impact. AI processes cross-market liquidity indicators to optimize order timing and venue selection. Only 11% of managers report AI as fully integrated in execution.

RISK MANAGEMENT & COMPLIANCE

AI generates risk hypotheses and identifies performance anomalies. VaR estimation benefits from generative adversarial networks. Compliance monitoring uses AI for screening, flagging, and real-time anomaly detection.

EVIDENCE OF ADOPTION: ROBO-ADVISORS, AI ETFS, AND PATENTS

LO 1: CURRENT AND FUTURE USES

Three data sources provide the clearest picture of the current adoption trajectory.

1) ROBO-ADVISORS: EXPLOSIVE GROWTH

AUM grew from near zero in 2017 to ~$1.8 trillion in 2024 (~4% of US equity market cap). Relies heavily on established ML, not sophisticated AI.

2) AI-DRIVEN ETFS: STILL TINY

AI-powered ETFs peaked at under $1B AUM in 2021 (<0.003% of US equity market cap). AI-driven strategies remain a very small market fraction.

3) PATENT EVIDENCE:

AI/ML patents in HFT rose to ~55-65% by 2022-23, up from near zero before 2015. Asset allocation patents with AI/ML exceeded 70% by 2022. Patent data serve as a forward-looking indicator of imminent integration.

WHERE AI CREATES THE MOST VALUE: ASSET CLASS RANKING

LO 1: CURRENT AND FUTURE USES

Strong positive correlation between liquidity and AI value creation score. Equities and commodities rank highest; infrastructure and private debt rank lowest.

WHY LIQUIDITY DRIVES AI ADOPTION

Real-time data, high volumes, dynamic prices, and transparency allow AI models to learn continuously. Equities and derivatives attract 57% of ML/GenAI adoption.

GENAI IN LESS-LIQUID MARKETS

GenAI can parse complex legal documents (e.g., bond indentures), lowering barriers in less-liquid markets like corporate bonds, private credit, and emerging markets.

HUMAN IN THE LOOP CONSENSUS & GENAI’S ROLE

LO 1: CURRENT AND FUTURE USES

A critical exam-relevant finding: strong consensus among market participants that full AI autonomy is not imminent. Understanding why this constraint exists is essential.

EVOLUTIONARY ADOPTION (NOW)

GenAI builds on existing methods:

• Writing and debugging trading code
• Summarizing research documents
• Extracting signals from LLMs for quantitative models
• Improving forecasting of textual data (e.g., central bank communications)

These are enhancements to existing workflows, not replacements.

REVOLUTIONARY ADOPTION (NOT YET)

Autonomous AI generating and executing trades without human oversight remains speculative.

AI-generated strategies that cannot be explained by humans are a “nonstarter.”

Regulatory, liability, risk management, and ethical constraints all reinforce need for human oversight.

THE 3-TO-5 YEAR HORIZON

Greater integration of sophisticated AI in investment and trading decisions expected.

Trend toward less human interaction, but complete autonomy is not anticipated.

Models will operate within predefined rules, with AI generating signals that humans or hybrid systems act on.

AI IS TRANSFORMING MARKET STRUCTURE: FOUR DYNAMICS

LO 2: MARKET DYNAMICS

GROWTH OF ALGO TRADING & NBFIs

• Algorithmic trading: ~70% of US equities, 50%+ of futures
• NBFIs hold over half of all financial assets globally
• AI gives agile NBFIs competitive edge over legacy-burdened banks
• Expected to expand across new asset classes and geographies

HIGHER SPEED OF MARKET REACTIONS

• LLMs enable real-time processing of unstructured data
• FOMC minutes: initial 0-45 second reaction now more accurately reflects eventual price impact
• Markets incorporate textual information faster and more precisely

HIGHER & MORE PROCYCLICAL VOLUMES

• AI-powered ETFs: portfolio turnover 10-20x higher than passive ETFs
• COVID-19 (Mar 2020): AI ETFs increased gross turnover as volatility spiked
• Higher turnover enhances price discovery normally, but amplifies instability under stress

HIGHER CORRELATIONS & INTERCONNECTEDNESS

• AI drives multi-asset, multi-venue strategies creating new correlations
• Firms using models trained on similar data from common vendors
• Portfolio responses to shocks may become synchronized
• Increased risk of simultaneous deleveraging and fire sales

ALGORITHMIC TRADING: A BALANCED ASSESSMENT

LO 2: MARKET DYNAMICS

AI will keep on advancing through algorithmic trading. The exam expects candidates to understand the dual nature of its market impact.

POSITIVE EFFECTS (NORMAL CONDITIONS)

• Liquidity improvement: Enhanced market-making capacity and tighter bid-ask spreads
• Informational efficiency: Rapid info incorporation, fewer price dislocations
• Reduced price impact: Execution algos minimize impact of large orders
• Price discovery: AI minimizes non-informational price swings

NEGATIVE EFFECTS (STRESS CONDITIONS)

• Flighty liquidity: Algos de-risk or shut down in high-volatility environments, making liquidity unreliable
• Volatility amplification: Correlated algorithmic responses to macro news amplify short-term volatility
• Flash crash risk: Cascading risk limits can cause sudden liquidity evaporation (e.g., US Treasury flash events)

FOUR FINANCIAL STABILITY RISK CATEGORIES

LO 3: FINANCIAL STABILITY

The IMF identifies four distinct risk categories. Candidates must describe and distinguish each, including transmission to the real economy.

Transmission channels to real economy: loss of market confidence, higher borrowing costs, and significant financial system outages

RISK 1: INCREASED MARKET SPEED AND VOLATILITY UNDER STRESS

LO 3: FINANCIAL STABILITY

This risk has three reinforcing mechanisms that candidates should be able to articulate clearly.

A: LEVERAGE & MARGIN AMPLIFICATION

• AI compresses bid-ask spreads and arbitrage margins
• Firms increase leverage to maintain returns
• Falling prices trigger margin calls, forced deleveraging, further declines
• AI-driven speed accelerates cycle vs. human portfolios

B: HERDING & CORRELATED MODEL RESPONSES

• Strategies trained on same data and open-source models produce similar responses
• Simultaneous rotation to safe assets creates self-fulfilling fire sale spiral
• IMF: herding was the #1 cited financial stability concern

C: MODEL SHUTDOWN & NOVEL EVENT FAILURE

• Risk limits tied to price signals shut models down during novel events
• Out-of-distribution events (e.g., COVID-19) produce incomprehensible outputs
• Humans must manually process high trade volumes, overwhelming capacity

RISK 2: OPACITY AND MONITORING CHALLENGES

LO 3: FINANCIAL STABILITY

AI creates new blind spots for regulators. The IMF identifies two main dimensions of this risk, both directly exam-relevant.

NBFI MIGRATION & REGULATORY ARBITRAGE

• Trading migrates from banks to NBFIs (hedge funds, prop firms)
• NBFIs face lighter model governance and explainability rules
• Sophisticated AI strategies migrate to less-regulated entities
• Systemic opacity increases, regulatory perimeter weakens

EMERGENT CROSS-ASSET CORRELATIONS and EMERGENT INTERACTIONS

• AI creates multi-asset, multi-venue, multi-geography portfolios
• New correlations invisible to single-institution monitoring
• Agent-to-agent trading and complex interactions between autonomous AI systems may produce behaviors impossible to anticipate from observing individual institutions
• IMF flags this as a source of emergent, unforeseen systemic risks

Key exam point: The problem is not just AI opacity internally, but that AI interactions across the financial system generate dynamics no single actor can observe.

RISK 3: OPERATIONAL RISK FROM THIRD-PARTY AI CONCENTRATION

LO 3: FINANCIAL STABILITY

Cloud infrastructure is dominated by 3 providers with 70%+ market share. Cloud disruptions totaled 205 hours in 2023 (up from 133 in 2022)

SYSTEMIC CONCENTRATION ARGUMENT

• High fixed costs create barriers favoring few dominant AI providers
• Smaller firms increasingly rely on third-party cloud/LLM services
• Single-provider failure affects multiple institutions simultaneously

KEY VULNERABILITIES

• Capital markets depend on real-time computation; latency = direct financial loss
• Smaller firms unable to build in-house AI are most dependent on third-party services
• Multi-cloud strategies are costly and complex to implement at scale
• Regulatory frameworks for critical third-party oversight are still nascent

EXAM POINT

IMF draws explicit analogy: AI provider failure comparable to failure of systemically important financial market infrastructure (e.g., central counterparties)

RISK 4: CYBER AND MARKET MANIPULATION RISKS

LO 3: FINANCIAL STABILITY

GenAI substantially lowers the cost and increases the sophistication of cyberattacks and market manipulation. The IMF notes these risks are already materializing, not merely prospective.

DEEPFAKES & SOCIAL ENGINEERING

• 2024: HK firm lost $25M via deepfake audio/video of senior executive
• GenAI enables highly personalized impersonation at scale
• Can trigger unauthorized transactions or manipulate asset prices

DISINFORMATION & SOCIAL MEDIA

• AI-generated content amplifies false information rapidly through social media
• Case: First Republic Bank targeted by online disinformation campaign
• Can trigger deposit runs or selloffs by weaponizing market sentiment

DATA POISONING & SYSTEMIC DISRUPTION

• Corrupted training data causes systematically wrong trading signals
• Affects all institutions using the compromised model
• 2023 ICBC ransomware attack disrupted US Treasury market conditions

AI is both the weapon and the shield in financial cybersecurity

THE REGULATORY LANDSCAPE: CURRENT STATE

LO 4: REGULATORY BEST PRACTICES

IMF surveyed 26 authorities in large capital markets. Existing frameworks are largely technology-neutral but important gaps remain for novel AI risks.

THE REGULATORY LANDSCAPE: CURRENT STATE

LO 4: REGULATORY BEST PRACTICES

KEY FINDINGS

• Standard-setters (FSB, IOSCO, BCBS): technology-neutral, results-based frameworks across 6 key areas
• National governments: 96% have data governance frameworks; 79% have national AI strategy
• Financial supervisors: mostly outreach (69%) and guidance (65%); only 4% enforcement

6 KEY REGULATORY AREAS

• AI governance
• Transparency & explainability
• Human-in-the-loop requirements
• Reporting, data quality & bias, outsourcing

EXAM TAKEAWAY

Supervisors are in a monitoring and guidance phase, not enforcement. IMF policy recommendations are forward-looking additions to this baseline.

AI FOR SUPERVISORS: SUPTECH

LO 4: REGULATORY BEST PRACTICES

The exam asks candidates to describe how supervisors themselves should use AI. The IMF identifies several high-value SupTech applications.

MARKET SURVEILLANCE & ANOMALY DETECTION

• Real-time monitoring of trading anomalies (price, volume, volatility)
• Earlier risk identification vs traditional reporting
• Flags misleading information and monitors transactions

REGULATORY COMPLIANCE & DATA QUALITY

• Automates data quality checks for completeness & consistency
• Resolves data linkage across sources without shared identifiers
• RegTech automates AML/KYC with higher accuracy

GENAI FOR SUPERVISORY EFFICIENCY

• Information retrieval, content creation, code generation & debugging
• Accelerates fraud detection & market monitoring
• Adoption gap: AEs 79% vs EMs 54% (needs upskilling)

SUPTECH: BEST PRACTICES FOR REGULATORS

midhafin.com

LO 4: REGULATOR BEST PRACTICES

BUILD CAPACITY

Invest in technical talent and infrastructure. Regulators must develop in-house AI expertise rather than relying solely on external vendors

MONITOR OUTCOMES

Establish clear frameworks for AI deployment in supervisory contexts. Define acceptable use boundaries, data quality standards, and validation requirements

SHARE KNOWLEDGE

Collaborate with industry and across jurisdictions. Pool resources to address common challenges in AI-driven financial supervision

MONITOR OUTCOMES

Continuously evaluate AI tools deployed for supervision. Track false positive rates, model drift, and effectiveness metrics over time

POLICY RECOMMENDATIONS

LO 4: REGULATOR BEST PRACTICES

International bodies (FSB, IOSCO) recommend a four-pillar approach to managing AI in financial markets:

1. Proportionate Regulation  —  Apply rules based on the materiality and risk profile of AI use, not blanket restrictions
2. Strengthen Governance Frameworks  —  Require boards and senior management to own AI outcomes with clear accountability chains
3. Monitor Macro-Prudential Impacts  —  Assess systemic risks arising from widespread AI adoption, including herding and concentration
4. Invest in Regulatory Technology  —  Equip supervisors with AI tools that can match the sophistication of the firms they regulate

IMF guidance: regulation should be balanced, technology-neutral, and principles-based rather than prescriptive

RECOMMENDATION 1:
ADDRESS MARKET SPEED AND VOLATILITY

LO 4: REGULATORY BEST PRACTICES

This recommendation targets the first financial stability risk category and involves both market microstructure tools and supervisory practices.

RECALIBRATE CIRCUIT BREAKERS

• Authorities and venues must assess if existing volatility mechanisms handle AI-driven price moves
• Circuit breakers may need re-parameterization for greater speed and amplitude
• Poorly designed breakers can worsen volatility and impede price discovery
• Testing algorithms in controlled environments before live deployment

REVIEW MARGINING PRACTICES

• Authorities, venues, and CCPs should review margin requirements for AI-driven price moves
• Foster preparedness for large variation margin calls during stress
• Assess margin model responsiveness to volatility
• Review initial margin in non-stress periods to reduce procyclicality

RECOMMENDATION 2:
ADDRESS OPACITY & MONITORING

LO 4: REGULATORY BEST PRACTICES

RISK MAPPING OF AI INTERDEPENDENCIES

• Require regulated entities to map interdependencies between data sources, AI models, and tech infrastructure
• Models may share common data sources, architectures, or third-party providers
• Current frameworks lack joint assessment of data dependencies
• Regularly updated risk map enables proactive management of systemic interconnections

ENHANCED NBFI OVERSIGHT & LARGE TRADER REPORTING

• Strengthen NBFI oversight with AI-relevant disclosure requirements
• Large traders must be uniquely identified and report activities to broker-dealers
• Enhance risk management and liquidity buffers at NBFIs
• NBFIs now hold over half of all financial market assets globally

RECOMMENDATIONS 3 & 4:
THIRD-PARTY RISK & OTC RESILIENCE

LO 4: REGULATORY BEST PRACTICES

REC 3: CRITICAL AI THIRD-PARTY PROVIDERS

• Adopt coordinated approach to defining and regulating critical AI service providers
• “Critical” must cover systemic AI model use, not just traditional IT infrastructure
• Interoperable regulatory approaches for cross-border compliance
• Require cybersecurity protocols at design, development, deployment, and operational stages

REC 4: OTC MARKET INTEGRITY & RESILIENCE

• OTC markets (fixed income, FX, derivatives) are less transparent and more fragile
• Collect and disseminate detailed OTC transaction data
• Require liquidity shift accounting in risk frameworks
• Expand market-maker incentives and central clearing for derivatives
• Backstop: central bank liquidity provision in shock events