Weekly

ISSUE 0066 November 2026
AI adoption in financial institutions

OUR MISSION

Helping your institution
succeed in AI transformation.

Frontwise helps your institution succeed in the AI era. We continually track AI developments in financial data, investment research and institutional organisations, sharing practices and perspectives worth your attention.

Let’s explore the possibilities of the AI era with a learning mindset.

Explore this week’s developments
Hong Kong financial district at sunrise; AI-generated illustrationAI-generated illustration

IN THIS ISSUE

Building authorization into products

10 financial AI cases and developments across data, investment research, organizations and regulation.

This week’s highlights

Financial data × AI

Tongdaxin packages data as MCP tools; Webull connects accounts to external assistants, with explicit execution boundaries.

Research & organizations

Samaya explores specialist training for lower costs; research benchmarks separate reports, forecasts and trading.

Morgan Stanley enables adviser-built agents; Aion and Osaic embed human approval in agent workflows.

China Universal brings fund services to Qwen’s general-purpose AI app.

Regulation × AI

The State Council report calls for exploring AI in financial management and services.

Black and white stones on a Go board; AI-generated illustrationAI-generated illustration
See the board. Shape the strategy.

Frontwise interprets financial AI trends to help you build on your professional strengths.

frontwise.ai

Financial data × AI

2 cases & developments

01

Tongdaxin × Huawei Cloud OfficeAce · Vendor article via ifeng · 2026-11-03

1. Tongdaxin brings 19 MCP tools to Huawei Cloud OfficeAce

Data
19 tools package prices, financials, flows and news across mainland, Hong Kong and US equities, funds, indices, futures, options and macro data
Coverage
100+ technical indicators, 15+ valuation models and 200+ candlestick patterns, plus supply-chain maps and option quotes
Workspace
Built on AgentArts, with three MCP connection methods and linked source citations
Boundary
The data layer supplies evidence; higher layers and users handle recommendations and decisions

The division of responsibility matters: evidence-backed data at one end and traceable citations at the other address two common points of failure.

This is a vendor-supplied article. Integration took place in October; capabilities are described by the partners.

Further reading: source ↗

02

Webull Canada · Investing.com report · 2026-10-27

2. Webull Canada connects brokerage accounts to AI assistants, with investor confirmation

Access
MCP connections to ChatGPT, Perplexity, Claude and Grok, without coding
Tasks
Check live prices, balances and holdings; analyze portfolios and prepare orders
Boundary
Investors must review and confirm orders in Webull; rollout to eligible accounts was expected in early November
Context
A similar US service preceded Canada; over 29 million global registered users, according to the company

Opening accounts to outside assistants goes beyond market-data access. Execution remains behind the broker’s own confirmation screen.

Based on Investing.com reporting; user numbers are company-reported.

Further reading: source ↗

OUR VIEWFrontwise view

Financial data and capabilities continue moving into MCP tools. Providers need to explain permissions, provenance and execution boundaries alongside coverage.

Investment research × AI

3 cases & developments

03

Samaya · Official technical blog · 2026-10-28

3. Samaya Sterling: a specialist model challenges frontier financial research performance

Results
58.9% of scoring criteria met on its own 220-question FrontierFinance benchmark, versus 55.9% for the comparator
Cost
$0.61 per task versus $6.89
Method
A 30B MoE model with 3B active parameters, two supervised fine-tuning stages and two reinforcement-learning stages, trained with 37 financial tools
Change
Median tool calls rose from 3 to 17 per task; the hardest group improved by 13.4 percentage points

Specialist training alongside tools offers institutions a route to lower research costs. Performance still needs testing on their own tasks.

Results come from the vendor’s own benchmark and have not been independently reproduced. Unverified comparator model names are omitted.

Further reading: source ↗

04

Prophetlab, Shanghai Jiao Tong University and partners · Reveneau report · 2026-10-27

4. Investor’s Last Exam separates research quality, forecasts and trading returns

Design
11 models and 86 equity-research tasks across discovery, evidence, forecasting, valuation, monitoring and synthesis
Corpus
A time-frozen 1.52TB corpus with 3.43 million document versions; only pre-cutoff data is available
Separate scores
Different models lead overall research quality, forecasting accuracy and simulated trading
Limit
Only 20 trading opportunities; this cannot establish real-world investment performance

A fluent report, an accurate forecast and a profitable trade require separate evaluation. Buyers need to define which capability the role requires.

Based on Reveneau’s coverage of the research and open-source project. Model rankings are omitted pending verification against the paper.

Further reading: source ↗

05

Taghash · entrackr report · 2026-11-02

5. Taghash launches Brain for fund reports and dashboards built on internal data

Capabilities
Portfolio analysis, reports, live dashboards and task execution across sourcing, portfolios, fund operations and LP management
Efficiency
The company says tasks that took days can take minutes
Adoption
The company says over 50% of active Indian VC funds use its platform
Roadmap
Builds on MCP integration and WhatsApp and Slack assistants, with further agent capabilities planned

Once data sits in a common system, fund software can let AI work directly on it. LP reporting and portfolio reviews are practical entry points.

Media coverage of the company announcement; efficiency and adoption figures are company claims.

Further reading: source ↗

OUR VIEWFrontwise view

Specialist training opens new cost options, while evaluations separate report quality, forecasting and trading. Buyers should define the task before choosing a score.

Organizations × AI

4 cases & developments

06

Morgan Stanley · American Banker interview · 2026-11-02

6. Morgan Stanley lets advisers build agents: almost 50,000 attempts, most ending in testing

Scale
Almost 50,000 attempts, including holdings tracking and daily summaries of large outflows
Review
Most end in early testing; adopted tools undergo accuracy and completeness review, with stricter checks for firmwide rollout
Internal service
AI handles roughly half of adviser inquiries first, routing unresolved questions to service teams
Quality
Over 1,000 domain experts evaluate output before production approval

The organizational question is how to decide which experiments merit further review. Actual adoption helps determine where deeper quality assurance is warranted.

An American Banker interview with the wealth-management AI lead. Planned capabilities include conversational portfolio analysis, risk simulations and proposal drafting.

Further reading: source ↗

07

Aion · Official release · 2026-11-02

7. Aion Agent handles business finance, with human approval for every payment

Tasks
Bills, invoice collection, bookkeeping updates and close reports across banking, payments, cards and credit
Boundary
Every payment, submission and signature requires human approval from the client team
Audience
Growing businesses seeking to reduce routine finance preparation

The agent prepares the work while people retain approval authority, addressing the daily time pressure on finance teams.

An official release, without independent usage data.

Further reading: source ↗

08

Osaic × Nitrogen · Joint release · 2026-11-03

8. Osaic approves Nitrogen’s full suite for nearly 10,000 financial professionals

Scope
Six products covering risk, research, income, tax, legacy and insurance
Agent capabilities
Nucleus turns statements into portfolios, prepares retirement-income presentations and visualizes tax insights, subject to adviser approval
Governance
Nitrogen claims to be the first wealthtech firm with ISO/IEC 42001 AI management-system certification

Large adviser networks consider both business coverage and compliance approval when buying AI. Governance certification is becoming part of the sales case.

A joint release; the “first” claim is company-reported.

Further reading: source ↗

09

China Universal Asset Management · China Fund News via Sina · 2026-11-02

9. China Universal’s wealth assistant launches inside Qwen

Format
An AI agent available through conversation inside the Qwen app
Capabilities
Fund comparisons, quarterly-report analysis, recurring-investment backtests and market summaries
Approach
Converts research, product analysis and investor-service experience into conversational assistance

Fund managers are placing expertise inside the general-purpose AI apps investors already use, changing how services reach them.

Media reporting on the company announcement, with no usage figures. The supplied draft dates the launch to 1 November.

Further reading: source ↗

OUR VIEWFrontwise view

What an agent can do depends on how precisely authority is defined. Employee building, agent preparation and system execution need distinct permissions and review.

Regulation × AI

1 cases & developments

Regulatory developments are included when relevant. This issue includes one item.

10

State Council financial-work report · Xinhua report · 2026-11-03

10. State Council financial-work report calls for exploring AI in financial management and services

Document
Submitted to the 25th session of the Standing Committee of the 14th National People’s Congress on 3 November
AI reference
Exploring AI in financial management and services appears in the section on strengthening financial regulation
Context
The report also addresses five priority areas of finance, technology finance and consumer protection

The placement matters: AI applications appear within the regulatory discussion. Implementation details depend on formal documents and subsequent policy.

Based on the Xinhua report cited in the supplied draft; not independently verified during this design work.

Further reading: source ↗

OUR VIEWFrontwise view

AI applications are entering the financial-regulation discussion. Institutions need to build responsibility, approval and audit trails into their workflows.

Editorial note

All cases and developments in this issue are drawn from the public sources linked above. Outcomes and capabilities disclosed by vendors and institutions are their own claims and have not been independently verified. Media-based items are identified individually. Frontwise views are editorial opinions, independent of the companies covered. This material is not investment advice. The supplied draft states an information-check date of 4 November 2026 and a next-issue date of 11 November 2026. Translation and layout follow that draft; no additional independent source verification is claimed.

Original shared draft ↗