Japan's First National AI Plan: Doubling Safety Staff and Loosening Data Rules for AI Training
December 12, 2025 - Tokyo. Japan's draft basic plan on AI lays out two immediate moves: expand the government's AI safety capacity and ease data-use rules to support model training. The Cabinet is expected to approve the plan within the month, marking the country's first formal, whole-of-government AI roadmap.
What's changing
The draft directs the AI Safety Institute, established in 2024, to double headcount from roughly 30 to about 60 "immediately." The institute will lead work on evaluating AI safety and related test frameworks.
To boost development, the government will submit an early bill to revise the Law on the Protection of Personal Information (APPI). The change would allow acquiring information without individual consent when the purpose is creating statistical data for AI training. For context on existing rules, see the Personal Information Protection Commission's legal overview: APPI (official).
The plan sets four principles: promote AI use, improve development capabilities, strengthen governance, and transform society. Managers across ministries will be urged to use AI in daily work to set the pace, and SMEs will get subsidies to adopt AI.
Why the push now
The draft is blunt: Japan is falling behind peers in AI development and investment, with direct consequences for industrial competitiveness and national security. The government intends to close that gap by leaning on a national strength-high-quality data-and by accelerating practical adoption across the public and private sectors.
Risk, trust, and oversight
The plan flags clear risks: criminal misuse, and the spread of misinformation and disinformation. It stresses the need to calm public concerns through governance and transparent evaluation. Scaling the AI Safety Institute is meant to speed up safety testing and establish consistent benchmarks.
No hard targets (yet)
Experts on the panel asked for numerical targets (citizen usage rates, total public-private investment). The government declined, saying it's too early to set those figures.
What this means for government, IT, and development teams
- Expect stronger internal signals to adopt AI, starting with leadership. Pilot tools with defined metrics and publish results to build trust.
- Prepare your data programs for a revised APPI. Even with broader allowances for statistical use, keep strict audit trails, access controls, and de-identification standards.
- Plan for safety evaluations. Log model versions, training data sources, red-team findings, and incident reports so you can respond to audits.
- For procurement, require vendors to disclose training data use, fine-tuning practices, evaluation methods, and update cadence.
- For SMEs, combine subsidy applications with a simple ROI model: a 90-day pilot, one high-impact workflow, a clear cost baseline, and weekly measurable outcomes.
- Upskill managers first. Adoption rises faster when leaders can set prompts, review outputs, and model responsible use.
Immediate actions to get ahead
- Inventory AI use cases with low risk and high value (document drafting, call summaries, coding assistance, analytics prototypes).
- Stand up a lightweight governance group (security, legal, data, operations) to approve pilots within two weeks.
- Draft a plain-language AI use policy covering acceptable data, review steps, logging, and human-in-the-loop checkpoints.
- Begin safety testing playbooks: prompt injection checks, toxic output filters, PII leakage tests, and hallucination baselines.
If your team needs structured upskilling and tool selection support, explore curated training by job role: AI courses by job.
Bottom line: the policy aims to speed adoption while strengthening oversight. Prepare your data governance, evaluation workflows, and leadership training now so you can move the moment the bill advances and the AI Safety Institute scales up.
Your membership also unlocks:
