Why I Want to Verify NRI's Expanded Anthropic Partnership and the Claude Cowork Financial Plugin
On February 24, 2026, Nomura Research Institute (NRI) announced the expansion of its partnership with Anthropic Japan.
NRI has introduced Claude for Enterprise across its entire company and is developing support services for the Japanese market. It is also proceeding with internal verification of Claude Cowork, a desktop AI agent.
On the same day, Anthropic released a financial plugin for Claude Cowork, which is also relevant to banking operations. Based on these developments, I will organize my expectations for the Claude Cowork financial plugin and outline scenarios for verification and implementation.
Note that while I primarily use Gemini for my daily work, this article was researched and written using Grok.
NRI's Announcement and Its Significance for Banks
In November 2025, NRI became the first certified Anthropic reseller for Amazon Bedrock in Japan. With this expansion, they are systematizing Claude implementation support for Japanese companies and strengthening their structure to provide end-to-end services from consulting to implementation and operation.
NRI itself has deployed Claude for Enterprise across its company within an AWS environment and is expanding its use in design, development, and consulting tasks. Regarding Claude Cowork, the company plans to verify the early evaluation version in its own operational environment and share the insights gained with Anthropic Japan.
Since it is common for banks to implement AI through system integrators (SIers), the fact that a major player like NRI is conducting its own verification indicates high reliability in terms of governance. From the perspective of model risk management and continuous monitoring required by the Financial Services Agency's supervisory guidelines, the increase in such partners is a welcome development.
Content of the Claude Cowork Financial Plugin
Claude Cowork is an agent that autonomously performs file operations and inter-application coordination on the desktop. The financial plugin added in this update integrates with institutional investor data platforms such as FactSet, LSEG, S&P Global, and MSCI. The main functions are as follows:
Financial Analysis: Market and competitor research, financial modeling (such as DCF), portfolio analysis, and automated workflows from Excel to PowerPoint.
Investment Banking: Transaction document review, comparable company analysis (comps), and pitch deck creation.
Private Equity and Wealth Management: Due diligence, target profile creation, and risk extraction.
These do not function as mere text generation but as structured workflows. Given that much of banking involves a combination of long documents, numerical data, and regulatory interpretation, efficiency gains in primary processing can be expected.
Claude's Features from a Banking Governance Perspective
The Financial Services Agency's comprehensive supervisory guidelines and AI discussion papers emphasize model risk management, accountability, and measures against hallucinations.
Claude's design is based on Constitutional AI, a mechanism that pre-embeds rules into the model to structurally suppress hallucinations. This is a strength, as it is easier to align with the continuous monitoring required by the Financial Services Agency.
It also has high accuracy in processing long documents, and stable logical development can be expected even when full supervisory guidelines, internal regulations, and past precedents are input at once. This has the potential to reduce risks caused by misinterpretation in compliance document reviews and credit screening.
Claude for Enterprise comes standard with centralized management, extended context, and audit logs, making it easier to comply with FISC security standards and the Act on the Protection of Personal Information. The fact that NRI is proceeding with its own verification also serves as a reference for banks when conducting PoCs.
Features from a Multi-Cloud Perspective
Anthropic provides Claude across all three major clouds—Google Cloud Vertex AI, Microsoft Azure Foundry, and AWS Bedrock, which serves as its primary foundation. This multi-cloud strategy offers the following advantages for banks:
Leveraging existing environments: If you are already using Gemini on Vertex AI, you can validate Claude in parallel without additional migration costs.
Risk diversification: You can secure alternatives in case of outages or price fluctuations without being dependent on a single cloud provider.
Flexible operations: Anthropic plans to make large-scale investments across all three cloud providers, ensuring high supply stability.
Since many banks operate in multi-cloud or hybrid environments, Claude can be considered a model with low barriers to entry.
Scenarios for Validation and Implementation
The following are specific scenarios where I want to validate the Claude Cowork financial plugin in banking operations, taking governance and multi-cloud into account.
Industry-specific screening
By inputting industry-specific screening criteria, financial statements, and industry reports, we can automate the extraction of risk points and reference similar precedents. This adheres to the point-based framework for industry-specific screening published by the Financial Services Agency and notifications regarding corporate value collateral rights. Claude supports this, allowing screening officers to focus on final decision-making. It can be operated in parallel with Gemini (for daily tasks) within Vertex AI.Transaction lending based on transaction data
By linking transaction history data and market data from FactSet/LSEG, we can support cash flow forecasting and collateral valuation. Utilizing Cowork's automated Excel-PowerPoint workflow streamlines the creation of screening documents. Audit logs allow for process tracking, making model risk assessment easier.Meeting minutes
By inputting meeting recordings or draft minutes, we can automate the extraction of key points, organization of action items, and proposals for follow-ups. Due to its high accuracy in processing long texts, we can expect precision from a compliance perspective. If we implement the organization of key points and action items according to meeting characteristics—such as for ALM committees or investment councils—we can also expect industry-wide expansion.Reputational risk including social media
By analyzing public social media data and news, we can support the flagging of reputational risks and impact assessment. Output with suppressed hallucinations prevents secondary risks caused by misinformation. Multi-cloud allows for flexible handling of data sources.Digital marketing
We can support campaign materials and customer segment analysis. It is possible to summarize market trends using data from sources like S&P Global and automatically generate PowerPoint materials. It is easy to combine with existing Google Workspace integrations.
The following are the estimated steps for implementation.
PoC contract for Claude for Enterprise via an SIer such as NRI
Building a small-scale RAG within Vertex AI or Bedrock (FSA guidelines PDF + precedent DB + internal manuals)
Validation of the Claude Cowork financial plugin (approximately 3 months)
Full-scale deployment after model risk assessment and audit log preparation
Costs will be equivalent to an enterprise plan plus SIer support fees, but these can be recovered through increased efficiency in screening operations and document creation.
Summary
NRI's announcement and the Claude Cowork financial plugin are directly linked to improving productivity in banking operations. If we leverage the flexibility of multi-cloud while prioritizing governance, the hurdles for validation can be kept low. Now that the Financial Services Agency is pointing out the risks of not taking on challenges, I believe it is worth paying close attention to these developments and testing them in the field.
(Engineers in the development department are using Claude, but I would like to use it on the business side as well.)
#GenerativeAI #Claude #BankingOperations #FSA #NRI #ClaudeCowork #MultiCloud #Governance