SYSTEM NOTICE

Auto translation by AI. Be sure, accuracy, nuances and authorial intent may not be fully reflected.

Domestic AI Agent Trends (February 24, 2026 Issue)

2026年2月24日 12:45

Update Date: 2026/2/24

Executive Summary
As of February 23, 2026, redesigns are progressing with a focus on 'closed environments, auditing, and operational visualization' for production deployment. AILEX launched an on-premises version and a multi-stage agent for digitized judicial procedures on the same day, ensuring accountability through PII masking and audit logs. Customer Cloud is providing a fully closed LLM for the pharmaceutical industry, and ExaWizards has expanded its municipal adoption to 53% nationwide. The challenge is a lack of context, making Agentic RAG the key, and integrated operations such as AgenticOps are urgently needed to prepare for the increase in shadow agents.

Infographic image of the article's overall picture created with Gemini 3

1️⃣ AILEX LLC: Two-part release of AI agents for LegalTech

📎 Source URL: https://www.value-press.com/pressrelease/370053 / https://www.value-press.com/pressrelease/370062
💡 Summary
AILEX LLC announced two releases on the same day. In the morning, they announced the launch of 'AILEX KAIKIRI,' an on-premises version that keeps data within the office server, adopting a configuration that sends data only after automatic PII masking when using external AI APIs (Claude, GPT-4o, Perplexity). In the evening, they added a multi-step agent for digitized judicial procedures (mints) that performs everything from OCR to AI classification, exhibit numbering, pre-submission checks, and package generation, along with features for detecting anomalies in client communication and automatic responses outside business hours. The design is characterized by achieving 'advanced automation' and 'accountability through audit logs' simultaneously.

2️⃣ Customer Cloud: Launch of fully closed local LLM for pharmaceutical and government agencies

📎 Source URL: https://prtimes.jp/main/html/rd/p/000000707.000099810.html
💡 Summary
Customer Cloud Inc. has begun providing a 'fully on-premises, closed-network' local LLM platform applying national-level data governance technology to the pharmaceutical industry. It enables AI analysis of research papers and clinical trial data without any data leaving the cloud. It is characterized by advancing data analysis and knowledge integration based on an 'AGI-driven development' framework, and it also integrates with BytePlus's AI cloud infrastructure. It is also attracting attention as a strategic piece of the Shibuya-based startup ecosystem 'Bit Valley 2.0'.

3️⃣ ExaWizards: exaBase Generative AI for Municipalities expands to 53% nationwide, 25 prefectural governments

📎 Source URL: https://exawizards.com/archives/31576/
💡 Summary
It has been revealed that 'exaBase Generative AI for Municipalities' provided by ExaWizards is being introduced and used in 25 prefectural governments, accounting for 53% of the country. While supporting the government-exclusive network (LGWAN), it comes with standard AI agent functions, allowing public servants to use agents for slide creation, minutes creation, legal research, and AI utilization proposals in a safe environment. With a track record of use in over 200 government agencies and municipalities, it is establishing its position as a major platform for administrative DX.

4️⃣ The 'Context Problem' of AI Agents: Agentic RAG is the key to practical application

📎 Source URL: https://www.itmedia.co.jp/business/articles/2602/24/news006.html / https://prtimes.jp/main/html/rd/p/000000063.000152797.html
💡 Summary
The 'lack of context (background information)' was raised again as the biggest barrier for AI agents to generate sustainable business value. The release of the 'SharePoint Data Management Course' by Useful Inc. (addressing the issue that employees spend an average of 65 minutes a day searching for internal information) is directly linked to the problem of corporate data structures where agents cannot reach necessary information. It is considered that the evolution from simple keyword-search RAG to 'Agentic RAG,' where agents actively acquire information based on reasoning plans, holds the key to practical application.

5️⃣ Rapid increase in shadow AI agents: 'Digital resilience' becomes a top management priority

📎 Source URL: https://ascii.jp/elem/000/004/375/4375737/
⚠️ Summary (Risk Alert Information)
According to Cisco and Splunk, 1.3 billion AI agents are projected to be operating on global networks by 2028, generating 394 zettabytes of data annually. Emphasizing that the increase in 'invisible AI' can become a management risk, they highlight the importance of visualizing and controlling which agents are accessing what data in real-time. Transitioning to a new operational model that detects and defends against AI agent runaway or breaches through 'AgenticOps' and integrated observability platforms via the integration of Cisco and Splunk is considered an urgent task.

6️⃣ Claude Cowork: Enters general phase for municipal document management in explanatory articles

📎 Source URL: https://www.itmedia.co.jp/business/articles/2602/24/news017.html
💡 Summary
ITmedia Business Online published an explanatory article introducing Anthropic's AI agent service 'Claude Cowork' in the context of municipal operations. Claude Cowork is a service where AI understands the content of files and folders specified on a PC and proposes and executes organization, analysis, and business automation. It is characterized by the ability to use AI with a more practical 'working together' feeling than conventional chatbots. Although it is an explanatory article rather than a primary announcement, it conveys the potential of AI agents for administrative DX in an easy-to-understand manner for general readers.

Comprehensive Review

In the legal field, a 'controlled hybrid' approach is becoming a realistic solution, where multi-stage workflows from OCR to submission generation are turned into agents, and external API usage is permitted with PII masking and audit logs. Medical and government agencies are prioritizing data sovereignty, moving forward with closed on-premises solutions, while municipalities are accelerating the horizontal deployment of platforms with standard functions compatible with LGWAN. On the other hand, the bottleneck for value creation is the scattering of internal information, and information design and permission control are prerequisites for Agentic RAG. Furthermore, the increase in 'invisible AI' is making operational visualization a management issue.

Future Points of Interest

A 'controlled hybrid' approach that does not rely solely on on-premises or closed environments, but also uses external models with PII masking and audit logs, is becoming mainstream. Procurement requirements standards will change.
For legal, the flow of OCR -> classification -> exhibit numbering -> pre-submission check -> package generation is a differentiator by being 'end-to-end until submission' rather than point automation. Standard metadata design is key.
For Agentic RAG, information structure is more effective than model performance. Companies that organize their SharePoint and other systems—including inventory, permissions, update responsibilities, and naming conventions—see greater growth in agent utilization.
Countermeasures for shadow AI agents are shifting toward an AgenticOps model that continuously cycles through detection, visualization, and control. The integration of observability infrastructure and SOC operations is the next point of contention.
As adoption grows among local governments due to LGWAN compatibility, the accountability for unauthorized access and erroneous generation during cross-departmental data linkage becomes heavier. Keep a close watch on trends in the standardization of logs and permission controls.