Japan's second AI Basic Plan (July 2026) explained in plain terms

Japan's second AI Basic Plan, adopted by the cabinet on 14 July 2026, is a government strategy, not a rulebook. It sets four priorities: adopt AI, build AI, govern AI and prepare society. It names compute, power and chips as national foundations. It leans on guidance and voluntary action rather than fines.

This post walks through the plan in plain language, using the official provisional English translation published by the Cabinet Office. It covers the law behind it, the four pillars, the infrastructure and power measures, the governance and cyber security measures, and the limits. Where a point comes from a news outlet or analyst and not from the plan, the post says so. Today is 19 October 2026, and the plan says it will be revised roughly every year.

For the physical build out of data centres, see Japan's AI data centres: power limits, Watt-Bit policy, new hubs. For the chip programme, see Rapidus explained.

Key numbers at a glance

• AI Promotion Act: Act No. 53 of 2025, passed by the upper house, the House of Councillors, on 28 May 2025 (The Japan News, via Asia News Network, 29 May 2025).
• First AI Basic Plan: cabinet decision on 23 December 2025 (Cabinet Office; Jiji Press via Adnkronos, 23 December 2025).
• Second AI Basic Plan: cabinet decision on 14 July 2026 (Cabinet Office).
• Guidelines on appropriateness of AI research, development and use: decided by the AI Strategic Headquarters on 19 December 2025 (Cabinet Office plan text).
• Government staff using the government AI environment called GENAI: 180,000 (second plan, 14 July 2026).
• Productivity effect: Japan's total factor productivity growth projected to rise by 0.2 percentage points as AI adoption accelerates (second plan footnote, citing a government council meeting of 24 June 2026).
• Chip related investment: public and private investment in semiconductors projected at 23.1 trillion yen, 10.5 trillion yen and 68.0 trillion yen by fiscal 2040, in a footnote that links them to vertical AI, physical AI and physical intelligent systems (second plan footnote).
• Plan revision cycle: in principle annually (second plan, Chapter 4).
• Personal information law amendment referenced in the plan: Act No. 56 of 2026.
• AI Robotics Strategy referenced in the plan: 26 March 2026.
• Council and expert panel supporting the AI Strategic Headquarters: both decided on 12 September 2025.

The law underneath: the AI Promotion Act

Japan did not write a detailed AI rulebook of the kind the European Union adopted. In May 2025 it passed a promotion law. The Japan News, republished by Asia News Network on 29 May 2025, reported that the upper house passed the bill with support from the Liberal Democratic Party, Komeito, the Constitutional Democratic Party, the Japan Innovation Party and the Democratic Party for the People. It described the law as the country's first on promoting generative AI development while managing its risks.

Key features reported at that time:

• A strategic headquarters, headed by the prime minister and made up of the whole cabinet, writes a basic plan for AI research, development and use.
• AI is defined as a technology that underpins economic and social development and also matters for security.
• The law includes measures to prevent the spread of false information and discrimination, and to respond to violations of citizens' rights.
• The government may investigate AI business operators and issue guidance based on what it finds.
• The government may disclose the names of companies that use generative AI for unscrupulous activities.

That report did not describe fines or a list of duties for businesses, and it did not state an effective date. The plan itself describes the approach as combining measures under the AI Act with related measures to encourage business operators and citizens to take voluntary and proactive steps. A regulation tracker, Regulations.AI, which is a secondary source, says the plan is a policy document that does not directly impose obligations or penalties on private companies. That matches the plan's tone, but it is the tracker's reading, and anyone with legal exposure should read the law in the original.

The basic plan is required by Article 18, paragraph 1 of the Act. The plan also refers to research and studies under Article 16 of the Act, which it says the government will run "in a more agile manner" to look at risks.

What the second plan is for

The first plan was adopted on 23 December 2025. According to Jiji Press, via Adnkronos, it set four fundamental policies: accelerate AI use, strengthen development capability, improve reliability, and keep transforming society with AI. It said Japan lags other advanced nations and that the gap is widening. The Daiwa Institute of Research, in an analysis dated 9 March 2026, agreed with the goals, such as models suited to Japanese culture and some autonomy. It also said it doubted the concrete measures, the capacity to execute and the speed.

The second plan, titled "Japan AI Transformation; A Stronger Nation. Greater Prosperity.", is a revision six months later. Its own text says the situation "has once again changed significantly" in that time. The headline shift is a focus on agentic AI, meaning AI that repeatedly makes plans, carries them out, checks the results and revises them, instead of only drafting text in a chat window. The plan says this is now spreading across industry, government, research and national security, and that how a society deploys and governs it affects economic and defence strength.

The plan also names the problems it wants AI to help with: population decline, insufficient domestic investment, stagnant wages and labour shortages in regions. It adds that gaps between Japan and other growing countries in AI use and private investment had widened by 2025, and it asks for AI to spread beyond large companies to mid sized firms, small businesses and local areas.

The four pillars

The outline lists four strategic pillars, with the working names "Adopt AI", "Create AI", "Enhance AI trustworthiness" and "Collaborate with AI".

Pillar 1: Adopt AI

This is about use, starting with government. The plan says the cabinet's own generative AI environment, GENAI, has been rolled out to government employees, mainly in central ministries, and that 180,000 government employees are using AI, which it calls the largest scale in Japan's public sector. It promises to expand GENAI to regional offices, to add agentic AI and Japanese made models, and to release it as open source so local governments can use it. It also commits the government to procure AI in a way that supports trust and transparency.

Beyond government, the plan says it wants "vertical AI", meaning AI built for one field and trained on that field's data, including tacit knowledge, the know how experienced staff hold in their heads. It names sectors including medical care, long term care, finance, education, disaster response, manufacturing, shipbuilding and logistics. It pairs this with "Regional AX", where AX means AI Transformation, the redesign of decisions and work around AI.

Pillar 2: Create AI

This is the infrastructure and development pillar. It uses the phrase "AI Sovereignty", which it defines as avoiding excessive dependence on any one country or company while staying open and working with like minded countries. The plan stresses strategic autonomy in government, defence and critical infrastructure, and says Japan will aim to be "indispensable" in semiconductors, data centres, communications, equipment and materials. It supports open source and open weight models, meaning models whose trained parameters are published, and calls for high quality Japanese language data.

It also emphasises "physical AI", meaning AI that understands the real world and takes physical actions, such as robots and autonomous driving. It refers to an AI Robotics Strategy dated 26 March 2026 and calls for domestic physical AI foundation models and stronger supply chains for components such as motors, speed reducers, sensors and storage batteries.

Pillar 3: Enhance AI trustworthiness

This is governance. The section on governance, AI safety and cyber security below covers it in detail.

Pillar 4: Collaborate with AI

This pillar covers work, schools and society. It asks for studies of AI's effect on employment, from both the substitution and the complement angle, and reskilling. It calls for studies of who is responsible when agentic AI causes harm, and it lays out five talent profiles: people able to use AI appropriately, AI researchers and developers, AI implementation professionals, AI governance professionals and AI innovation professionals. A footnote notes that the plan is discussing the difference between "human in the loop", where a person takes part in each task, and "human on the loop", where a person designs and monitors the process while AI does the work. It also asks for revised guidelines on generative AI use in schools, and for a look at how to protect intellectual property while still allowing AI use, including fair value return to rights holders.

Compute, chips and power

For readers following the infrastructure story, the plan is broad and short on numbers. It lists these measures in its infrastructure section:

• Support for data centres, compute for AI research, and a data and communications backbone, including what it calls Watt-Bit Collaboration, the planning of power and network capacity together, plus an All Photonics Network and next generation communications research.
• Research and development of highly capable frontier AI semiconductors, led by the trade ministry.
• A next generation flagship system after the Fugaku supercomputer, and wider shared compute for researchers.
• Expansion of domestic production and supply capacity for data, cloud, data centres, computing platforms and semiconductors, with a commitment to review electricity demand forecasts regularly and ensure stable power supply.
• Strategic domestic siting of data centres, including through the GX Strategic Area system, and exploring overseas expansion of the data centre ecosystem. GX stands for green transformation, Japan's name for its decarbonisation industrial policy.

The one hard set of numbers is a footnote. It says that at a government council meeting on 24 June 2026, Japan's productivity growth was projected to rise by 0.2 percentage points as AI use accelerates, and that public and private investment in semiconductors for vertical AI, physical AI and physical intelligent systems was projected to reach 23.1 trillion yen, 10.5 trillion yen and 68.0 trillion yen by fiscal 2040. The wording in the provisional translation is ambiguous about whether the three figures are separate categories or something else, so check the Japanese original before quoting them. They are government projections, not commitments of spending.

The plan itself carries no budget table. Jiji Press reported in December 2025 that the government hoped to compile a roadmap with an investment target, and the Daiwa analysis said it was due by spring 2026. No roadmap figure had been published in the official plan documents as of 19 October 2026, so that remains an open question.

Governance, AI safety and cyber security

The governance section is the one security teams should read. It makes these points:

• Guidelines on ensuring appropriateness in AI research, development and use were decided by the AI Strategic Headquarters on 19 December 2025, and the plan says other guidelines have been prepared to match.
• The AI Safety Institute, known as AISI, which handles technical evaluation of AI models, is to be strengthened, including through more staff with technical expertise, and its results are to be connected to ministry rules and operations. AISI is to work with counterparts through an international network that includes the United States CAISI and the United Kingdom AISI.
• Japan will promote the Hiroshima AI Process, an international framework it led, and will aim to hold an AI Summit in Japan "at an early date."
• The government plans to expand public consultation services for people harmed by AI and to help companies set up complaint handling.
• It will support technology to tell AI generated content apart, such as digital watermarks.

The plan is direct about security. It says the risks of agentic AI include large failures from chains of malfunctions, a blurring of who is responsible, and attacks in which AI autonomously discovers system weaknesses, builds attack steps, runs them and adjusts. In response it describes a cyber security package called Project YATA-Shield. Under it, the government will assess weaknesses in critical government systems using highly capable frontier AI, issue alerts to operators of critical infrastructure, and issue alerts to software vendors about early discovery and response to vulnerabilities. AISI is to collect information and evaluate the cyber security performance of AI models. The plan also tells organisations that managing cyber risk should happen under the leadership of executive management, and that they should make active use of highly capable AI to speed up the discovery and correction of vulnerabilities.

None of that is a legal requirement on a company. It is a statement of government direction. But a software vendor that sells into Japan should expect questions about how quickly it finds and fixes vulnerabilities, and about how it governs its own use of AI.

Data and privacy

The plan also mentions an amendment to the personal information law, Act No. 56 of 2026. It says the amendment includes special provisions that remove the need for a data subject's consent when personal data is provided to third parties solely for AI development and other activity classed as statistical compilation, together with a penalty system for breaches of those provisions. The Personal Information Protection Commission is to draft the subordinate rules. This is the one place in the plan where penalties appear, and it is in a separate law that the plan points to, not in the AI Promotion Act. The plan summarises the amendment rather than reproducing it, so the details should be checked in the amended law itself.

The plan also calls for measures against leakage of sensitive information, including trade secrets, overseas, as part of its data strategy.

What the plan does not do

• It does not set a budget. The plan lists ministries responsible for each measure and leaves funding to budgets and other documents.
• It does not impose fines on companies. The plan describes no penalties of its own, and the AI Promotion Act, as reported in 2025, allows investigation, guidance and disclosure of names.
• It does not copy the European approach. The plan says Japan will take account of developments in other countries, and the Daiwa analysis notes that the United States and the European Union prioritise different items and use different methods.
• It does not remove uncertainty about execution. The Daiwa Institute raised that doubt about the first plan in March 2026.

What is firm and what is uncertain

Firm, because the plan itself states it: the adoption dates, the four principles and four pillars, the GENAI scale of 180,000 users, the AISI strengthening, the YATA-Shield measures and the annual revision cycle.

Uncertain: how much money will follow, whether the investment roadmap has been published, how much power capacity the data centre measures will unlock, how strongly the government will use its investigation and disclosure powers, and how the plan will be revised next year. The English text is marked as a provisional translation and says to refer to the original for accuracy.

What to watch next

• The next annual revision of the AI Basic Plan.
• A published public private investment roadmap with figures.
• Draft rules under the personal information amendment.
• AISI evaluation outputs and staffing.
• Any use of the investigation and name disclosure powers.
• The date and agenda of the AI Summit in Japan.

For vendors selling into Japan, Safeguard offers SBOM and dependency scanning, with a free scan at https://gold.safeguard.sh/scan.

Quick check

Sources

• Cabinet Office of Japan, Second Artificial Intelligence Basic Plan, 14 July 2026, provisional English translation: https://www8.cao.go.jp/cstp/ai/ai_plan/aiplan_eng_20260714.pdf
• Cabinet Office of Japan, Summary of the Second AI Basic Plan, 14 July 2026: https://www8.cao.go.jp/cstp/ai/ai_plan/aiplan_g_eng_20260714.pdf
• Cabinet Office of Japan, AI Basic Plan page listing both plans: https://www8.cao.go.jp/cstp/ai/ai_plan/ai_plan.html
• The Japan News via Asia News Network, House of Councillors passes AI promotion bill, 29 May 2025: https://asianews.network/japans-house-of-councillors-passes-ai-promotion-bill/
• Jiji Press via Adnkronos, Japan government adopts first basic plan on AI, 23 December 2025: https://english.adnkronos.com/2025/12/23/japan-govt-adopts-1st-basic-plan-on-ai/
• Daiwa Institute of Research, Trustworthy AI is the goal in the AI Basic Plan, 9 March 2026: https://www.dir.co.jp/english/research/report/analysis/20260309_025618.html
• Regulations.AI, Japan's AI Basic Plan (secondary tracker): https://regulations.ai/regulations/RAI-JP-NA-JAPANBA-2025

Last reviewed: 19 October 2026.