Weekly

Morningstar Direct AI and MSCI SignalLab: data providers turn content into agents on the same day

Morningstar Direct AI

Three agents for product development (category selection), distribution (winning portfolio placement) and manager research (due diligence), available globally at launch

Data distribution

Morningstar data also connects to Claude, Microsoft Copilot and ChatGPT, allowing clients to use it without opening Direct

Foundation

Millions of securities and entities, backed by the ratings and research frameworks of more than 1,100 analysts

MSCI SignalLab

Released the same day, with over 600 research-grade signals updated daily. In Risk Premia, AI reads economic and financial literature to generate candidates that MSCI researchers validate individually

Morningstar’s agents address specific decisions. Product development combines flows and categories to explore demand, competition and comparable products’ growth. Distribution starts with a target model portfolio, assesses gaps and the effect of adding a fund, then prepares meeting points. Manager research applies the firm’s analyst framework to changes in trading behaviour and performance across market conditions.

MSCI SignalLab standardises signal definitions and documentation across Risk Premia, Micro-Industries, FactorLab and Crowding. Its Risk Premia agent turns literature into tested, scored candidates for researcher validation. Micro-Industries uses language models for company clustering, classification definitions and stock assignments, followed by model-team validation.

Both embed research methods into products. Morningstar also distributes data through general-purpose assistants; MSCI supports proprietary-data integration and access through API and Snowflake. These are official product descriptions, not demonstrated investment outcomes.

Further reading: Morningstar official release (English) ↗
Further reading: MSCI SignalLab official release (English) ↗

LSEG: GPU acceleration brings analysis of 110 trillion Tick History rows down to seconds

Data scale

110 trillion rows covering 100 million instruments, more than 30 years and over 580 exchanges and venues

Acceleration

Vendor benchmarks report that NVIDIA cuDF and Polars reduce analysis of multiple years of data from hours to seconds

Permissions

No additional complete copy of the data; existing licences, permissions and access controls are retained

Demand

Real-time network traffic rose 29% year on year in H1 2026, peaking at 26 million messages per second in June; Tick History API requests increased 48%

The collaboration changes how historical market data is processed. NVIDIA cuDF and Polars accelerate work on permissioned Tick History data without requiring another complete copy. Existing licensing and access controls remain applicable, while retaining granular information rather than downsampling it away.

LSEG identifies three workflows: testing strategies across longer histories and more venues; reconstructing order books and assessing prices, liquidity and execution; and accelerating scenarios and market replay so clients can apply their own analytical or AI models to governed data.

Near-continuous trading makes overnight analysis less convenient. The aim is to bring historical analysis closer to intraday decisions. Hours-to-seconds figures are vendor benchmarks; the release does not specify a complete hardware or task configuration or provide customer measurements.

Further reading: LSEG press release (English)

Omega Point: Anthropic-verified portfolio-analysis MCP enters Claude’s marketplace

Verification

All connectors in Claude’s marketplace undergo automated checks; “Verified” indicates direct assessment of quality and security by Anthropic

Usage history

Omega Point says investment teams have used its MCP in production within Claude since May this year

Computation

Factor exposures, what-if analysis, stress tests and attribution are calculated in real time using clients’ holdings, permissions and data, on the same engine as their formal risk reports

Claude supplies the conversational interface while Omega Point’s deterministic engine performs the calculations. The vendor stresses consistency with the models behind an institution’s formal risk reports, avoiding a separate approximation inside the assistant.

Examples include examining sector-level factor risk for a portfolio sleeve, identifying names contributing a tenth of risk alongside month-to-date profit and loss, and pasting a proposed trade list to compare exposures and stress results before and after trading. Teams can also load a complete holdings file.

Proprietary research and signals can enter the same environment, with answers limited to the user’s existing permissions. Anthropic verification concerns connector quality and security; it is not endorsement of investment conclusions or a substitute for institutional checks.

Further reading: Omega Point press release (English)

Securities Association of China: over 70% of brokerage research departments use AI as report volume declines

AI adoption

Over 70% of research departments used AI in 2025: basic uses included information gathering, proofreading and transcription; deeper deployments included locally hosted large models and proprietary research systems

People

100 brokerages had research departments, employing 7,591 research staff (down 3.68% year on year), including 5,886 analysts (up 4.38%)

Output

98 brokerages published 205,503 reports during the year, down 11.54%

Revenue

Institutional-client trading commissions totalled RMB15.894 billion, accounting for 85.98% of research departments’ total revenue

The figures also describe changes in staffing and economics. Analysts represented 77.54% of research staff; departures reached 1,972 against 1,644 new hires. Report output fell as business concentration rose, suggesting that volume alone is an incomplete measure of a department’s direction.

Basic AI uses cover gathering information, proofreading, formatting, transcription and editing. Deeper uses include locally hosted models, proprietary research systems, research methods, thematic analysis and compliance. The adoption percentage measures institutions using AI, not the share of tasks automated.

Internal collaboration generated 3.57% of revenue, while other income, mainly consulting, contributed 7.29%. The association links service transformation with valuation methods, expert-fee controls and information-source management. These are industry statistics reported by the media, not an evaluation of AI returns.

Further reading: Securities Daily report (Chinese)

Large models underpin brokerage wealth management: from individual tools to end-to-end infrastructure

Sample

A joint series by Cailian Press and the association, with more than 30 published brokerage-executive interviews

In-house development

East Money Securities’ Miaoxiang model supports report interpretation and portfolio diagnosis. CICC Wealth has four agent groups covering adviser assistance, research, wealth-management digital employees and client-facing services

Co-development

Zhongtai Securities and Alibaba Cloud jointly developed the Qianwen Qiantai advisory brand, reporting more than 700 service scenarios

Expanding scope

AI extends from research and advice into compliance: pre-service suitability screening, real-time monitoring of improper stock recommendations and subsequent end-to-end audit trails

Industry base

Almost 90,000 registered investment advisers across the industry

The review follows research into account-level service. East Money combines report interpretation, news analysis and portfolio diagnosis, with plans for account-level services and external agents. Zhongtai combines its own infrastructure with cloud co-development; CICC Wealth groups agents around advisers, research, internal employees and clients.

Guojin places AI across pre-investment assessment, ongoing monitoring and reviews, leaving individual needs to advisers. Huayuan assigns standard processing and account monitoring to technology, with experts handling complex judgement and risk communication. Minsheng identifies front-line staff and advisers as the immediate users.

BOC International (China) describes suitability screening before service, communication monitoring during it, and electronic records afterwards. Longer-term retention and client-experience incentives accompany the technology. These are institutional accounts of operating practices, rather than independently tested results.

Further reading: Cailian Press report (Chinese)

Asset-owner survey: AI becomes a top strategic priority while governance lags

Sample

20 global asset owners managing more than US$5 trillion in total

Priority

67% named AI, including agentic tools, as a top strategic priority, ahead of inflation (44%), liquidity pressure (33%) and private markets (28%)

Expectations

Technology’s contribution to investment outcomes is expected to rise from about 17% today to 30% within three to five years

Governance gap

Only 25% include AI as a standing board or investment-committee agenda item; 60% review it only occasionally

Main constraint

Three quarters of respondents identified data quality and standardisation as the biggest obstacle

The study places AI inside research, portfolio monitoring and decisions. Participants recognise its importance but report that skills, systems and operating models remain unprepared. Human interpretation and accountability remain necessary as investment and technology teams reorganise work.

Legacy systems, organisational silos and spreadsheets constrain trusted data access. External providers can fill capability gaps, although dependence on a few platforms could increase convergence risk.

The report proposes four priorities: clear leadership ownership; people who bridge investment and technology; redesigned workflows preserving human judgement; and sustained investment with value measures. It addresses organisational capability rather than additional tool purchases.

The 20-owner sample is not a global census. The 67% priority figure concerns participants disclosing that measure; the 25% agenda figure measures governance and cannot be subtracted from unrelated adoption rates.

AI is becoming an organisational story rather than a technology story.

Roger Urwin, co-founder of the Thinking Ahead Institute (official press release)

Further reading: Thinking Ahead Institute press release (English)

Itaú: ia.i targets all clients; IARA governs models, costs and permissions

ia.i

A unified conversational layer across the bank, using text or voice for accounts, credit, cards and insurance; initially around 300,000 clients, targeting all roughly 70 million by the end of 2026

IARA

An internal governance and orchestration platform that grants teams model access under policy, manages token costs and observability, and implements responsible-AI requirements

Legal use case

A library of more than 90 million documents, over 2 million tasks per month and reported efficiency improvements of 20–80% in selected processes

Internal targets

Three times as much time for client relationships and twice the service speed (targets)

ia.i’s customer experience and IARA’s governance layer reinforce each other. Text and voice interactions can connect to specialist human support. Customer DNA supplies governed financial context about clients’ habits and life stages so services can become more relevant.

IARA controls model access, observability, token costs and responsible-AI policies. Model selection reflects task complexity and latency requirements. The legal-document application provides a concrete large-scale use case, although its reported gains apply to selected processes rather than the whole bank.

Relationship managers also receive meeting preparation, summaries and alternative plans. The initial customer group supports learning from actual behaviour. Full-base access, product simulations and recurring financial analysis are rollout goals; relationship-time and service-speed improvements remain expectations.

The challenge is to make technology disappear: simple, direct and useful, delivering immediate benefits to clients while remaining almost unnoticed.

Carlos Eduardo Mazzei, Itaú CTO (as attributed in the supplied OpenAI case; wording translated from Chinese)

Further reading: OpenAI customer case (English)

Mizuho: mascot Aomaru becomes a conversational AI agent, initially without account access

Positioning

The official mascot Aomaru becomes the AI agent Aomaru Bank, initially answering questions about banking services and procedures

Knowledge base

More than 1,700 official FAQs and website information underpin retrieval. Answers cite sources and direct users to a human telephone service when needed

Voice

Real-time voice conversation is supported at launch using OpenAI voice technology

Safety design

Multiple layers of input and output checks and controls; no account-information or transaction-history access initially; client data is not used to train OpenAI models

Aomaru Bank turns FAQ searching into a guided conversation. It clarifies a person’s situation, organises official information, cites the underlying pages and offers a human channel when necessary. Its initial role is guidance rather than autonomous account operation.

Real-time speech combines OpenAI voice technology with Mizuho’s agent functions. On-screen transcripts let customers check responses, including those less comfortable typing. OpenAI engineers provided technical support and advice during development.

Input and output controls address inappropriate content and malicious instructions. Initial account and transaction access is excluded; conversation analysis is limited to service improvement and data is not used for OpenAI training. Expansion into procedures or asset-building advice depends on feedback and regulatory considerations. No usage or accuracy figures are disclosed.

Further reading: Mizuho press release (Japanese)

Huntington: an AI operating system for small-business owners, including non-customers

Product

The 1866 Studio, a small-business operating system built on Google Cloud’s Gemini platform

Scenarios

Starting a business (registration, permits and action plans); multimodal generation of marketing materials and advertising copy; loan-readiness assessment and improvement suggestions

Access strategy

Early registration opened on 19 October, explicitly including people who are not Huntington customers

The 1866 Studio combines banking expertise, an AI assistant, expert content and operating tools. Its intended use of business context across tasks helps owners coordinate finance and operations while retaining decision control.

Initial workflows cover entity selection, registration and launch planning; marketing materials, advertising and social campaigns; and financing readiness. The latter identifies gaps and explains lender expectations—it is not a credit commitment or a replacement for underwriting.

The roadmap extends across business formation, operations, growth and eventual succession or exit. The 19 October announcement opened early-access registration, including non-customers; it does not establish broad deployment or measured business outcomes.

Further reading: Huntington press release (English)

MAS: AI risk-management guidelines for financial institutions, phased through 2028

Four expectations

Board and senior-management oversight and accountability; lifecycle risk management (inventories, risk tiers, data governance, testing, human oversight, cybersecurity, monitoring and change control); third-party AI risk management; and proportionate application

Implementation

Effective 7 October 2027: the oversight and management-system sections must be met. Lifecycle controls and capability-building sections may be implemented by 7 October 2028

Agentic AI

MAS says it will consult the industry in 2027 on additional guidance for agentic AI

Third parties

Institutions retain responsibility for externally provided AI; services should be restricted, suspended or replaced when risks cannot be brought within appetite

MAS addresses both enterprise-wide oversight and individual use cases. Institutions should inventory AI at suitable granularity, assess materiality and match controls to risk across testing, data, human oversight, cybersecurity, monitoring and change management. Embedded AI still needs attention.

For third-party services, institutions should obtain assurance, test suitability and use compensating controls where evidence or access is limited. Services may need restriction, suspension or replacement when residual risk exceeds appetite. Outsourcing the model does not outsource accountability.

Low-impact uses may need basic procedures, while higher risks require stronger controls. Existing committees can provide adequate oversight. Sections 3–4 apply from October 2027, with Sections 5–6 by October 2028; further agentic-AI consultation remains planned for 2027.

Further reading: MAS official media release (English)