AI agent training for IT administrators in public sector in Japan

Japanese ministries and local governments need IT administrators trained to connect AI agents to standardised administrative systems, keep national identification number data out of reach, and route held actions through the existing approval chain. Privileged access, connections and incident decisions stay with them. The Act on the Protection of Personal Information (APPI) is the central data law.

What do IT administrators in Japanese public bodies handle?

Local government IT has been going through a large migration. Core systems for resident registration, taxes, national health insurance, childcare and welfare are being moved to standardised specifications on a shared government cloud, replacing many locally customised systems. Administrators in a city office spend much of their time on that migration: data conversion, testing, and keeping old and new systems running side by side.

Alongside it sit the usual duties: accounts for staff who rotate between departments every few years, network separation between systems that handle the national identification number and those that connect to the internet, endpoint management, and supplier coordination. National ministries add their own mission systems and higher security classifications. Japanese administrative work also runs on formal approval chains, where a document passes up through several levels before it takes effect.

Where do agents help, and at what depth?

Routine IT tasks suit agents well, provided each runs at a depth the body has proven on its own cases:

• Migration data checks (depth 1). The agent compares records converted to the standard specification with the source system and lists mismatches for staff to fix.
• Staff rotation access changes (depth 2). When personnel orders move staff between departments, it drafts the access changes. An administrator approves.
• Helpdesk triage (depth 2). It classifies staff enquiries and drafts answers in Japanese from the internal manual.
• Patch and maintenance notices (depth 3 within limits). It notifies departments of planned work and logs confirmations.

Administrators also connect business agents, for example one that drafts replies to residents about procedures, and decide what each may read. Our piece on which roles in the public sector need AI agent training explains why IT staff need it early.

What must IT administrators learn, and what stays with them?

First, the approval chain. An agent's held action waits for a named person. In a Japanese office, that person should be the one the existing approval rules designate for that kind of decision, not whoever administers the system. Administrators configure where held actions go, so they need to map the agent's verdicts onto the office's approval chain. The way payroll staff handle held items shows the pattern; our guide on how payroll administrators handle an action an AI agent has held walks through one.

Second, separation. Systems that handle the national identification number sit on separated networks with strict limits on use. An agent should not be connected across that separation for convenience. Every action meets five bounds, role scope, certificate, sources, the body's own limits and the platform guardrails, and administrators control role scope directly.

Third, certification. Depth is earned on the body's own cases through Attest, and agent certificates lapse after 90 days; a run that clears no depth drops the agent to depth 1 the same day. Administrators should set alerts and know what to do when an agent's depth falls.

What stays with the administrator: privileged access, any connection to separated networks, approval of new connections to resident data, incident declaration, and supplier access.

Which Japanese rules shape oversight and literacy?

The APPI governs personal information and now sets common rules for private businesses, national administrative organs and local governments, with the national personal information regulator overseeing them. It requires specified purposes, appropriate security management, supervision of contractors and staff, and breach reporting. Our piece on the APPI and training staff who use AI agents covers its training implications. Use of the national identification number is governed by separate rules that restrict it to defined administrative purposes.

Government guidance on generative AI expects ministries to manage risks, protect confidential information and keep people responsible for outputs. The FSA supervises financial institutions and does not reach a city office. As a benchmark for what staff literacy should cover, the EU AI Act's Article 4 for the public sector is a useful comparison, though it does not apply in Japan.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

A 90-day training plan

Days 1 to 30. IT staff learn depth, the minimum chain, provenance and verdicts. The IT section chief inventories agents and maps each one's held actions to the approval chain.

Days 31 to 60. Two administrators pilot migration data checks at depth 1 and access changes at depth 2. They rehearse a scenario where an agent's certificate lapses mid-migration.

Days 61 to 90. Confirm no agent crosses network separation, document each agent's scope for the personal information officer, and set certificate expiry alerts.

Where the Academy fits

Pryme Intelligence Academy's Foundations track gives IT staff the vocabulary free, and its lessons can be listened to as audio with highlights and notes. Administrators who run the workspace take the Operator track, whose module tests are set in the public sector and the IT role. Our view is that the public sector will not scale AI agents without trained reviewers, and IT is part of that.