ใƒกใ‚คใƒณใ‚ณใƒณใƒ†ใƒณใƒ„ใธใ‚นใ‚ญใƒƒใƒ—
่ฆ‹ๅ‡บใ—็”ปๅƒ

[Series #2] Governments Are Deployers โ€” And the EU AI Act Holds Them to the Same Legal Standard as Corporations

    โ–Œ The Legal Innovation at the Heart of the EU AI Act

    The EU AI Act did something no prior AI framework had attempted: it named governments as regulated parties.

    Not merely as rule-setters or supervisors โ€” but as deployers, with the same binding legal obligations as any corporation that puts a high-risk AI system into use for professional purposes.

    This is the concept that separates the EU AI Act from every other AI governance framework in the world. And its implications for how we evaluate government AI adoption in Japan, the US, and EU member states are profound.

    โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€

    โ–Œ Article 26: What "Deployer" Means in Law

    Article 3(4) of the EU AI Act defines a deployer as "any natural or legal person, public authority, agency or other body that uses an AI system under its own responsibility." The key phrase is "under its own responsibility." The moment your organisation deploys an AI system for professional purposes, you are a deployer โ€” regardless of whether you built the system.

    For public authorities deploying AI in citizen-facing services, the operational obligations under Article 26 are not aspirational. Public authorities deploying high-risk AI systems must comply with registration obligations, ensure the system is registered in the EU database, assign human oversight to competent and trained individuals, retain automatically generated logs for at least six months, inform individuals subject to AI-assisted decisions, and cooperate with authorities on regulatory actions.

    A public authority using AI in benefits or employment decisions is โ€” unsurprisingly โ€” also a deployer. And critically: these obligations cannot be shifted to the vendor by contract. The compliance gap belongs to the deployer.

    Additionally, a Fundamental Rights Impact Assessment (FRIA) applies to public bodies and private entities providing public services that deploy Annex III systems. This is the AI equivalent of a GDPR Data Protection Impact Assessment โ€” a structured, documented pre-deployment assessment of how the AI system may affect the fundamental rights of the individuals it touches.

    โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€

    โ–Œ The Stakes: โ‚ฌ15M or 3% of Global Turnover

    Penalties for deployer non-compliance reach up to โ‚ฌ15 million or 3% of worldwide annual turnover โ€” whichever is higher. For mid-sized and large enterprises and public authorities, that is not a fine that rounds to zero.

    The high-risk Annex III categories that most directly implicate government services include: employment and HR management, access to essential services and public services, law enforcement, border control management, administration of justice, democratic processes, and education.

    In short: the most consequential things governments do with AI are precisely the things the EU AI Act regulates most stringently.

    โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€

    โ–Œ Japan: The World's Most Ambitious Government Deployer โ€” With No Equivalent Obligation

    Japan is, by any measure, executing the most ambitious government AI deployment programme in the world right now.

    Japan's Digital Agency is running "Government AI GENNAI," a large-scale pilot targeting approximately 180,000 government employees across all ministries and agencies in fiscal year 2026. The platform launched broader rollout in early 2026 and reached over 100,000 officials by May 2026.

    Each ministry is required to appoint a Chief AI Officer (CAIO), and the Digital Agency's procurement guidelines include governance checkpoints, contract checklists, and explicit risk review logic โ€” including consideration of whether data could be censored or accessed by foreign governments where systems rely on servers outside Japan.

    This is thoughtful, sophisticated governance. But it is soft law. Japan's AI Promotion Act imposes no fines, no mandatory FRIA, no obligation to register AI systems in a public database, and no structured incident reporting to regulatory authorities.

    Gennai's security, auditability, and logging controls function as de facto baseline expectations for vendors and contractors โ€” but they are operationally enforced norms, not statutory rights for citizens.

    Measured against the EU AI Act's deployer standard: Japan's government is deploying at scale without the legal accountability architecture that the EU now requires.

    โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€

    โ–Œ The United States: Deployer Without Doctrine

    In May 2026, the Department of Defense announced agreements with eight leading AI companies to deploy their capabilities on classified networks, establishing what the administration described as the premier AI-enabled fighting force.

    There is no US federal equivalent to the FRIA. There is no statutory obligation for federal agencies deploying AI in citizen-facing services to maintain logs, conduct conformity assessments, or register systems in a public database. The June 2, 2026 EO builds a voluntary cybersecurity framework โ€” not a rights-protection framework.

    The US government is a deployer by function. It is not a deployer by law.

    โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€

    โ–Œ Three Strategic Implications for Leaders

    โ‘  The providerโ€“deployer boundary matters more than most organisations realise. If your organisation substantially modifies a third-party AI system and places it on the market under its own name, you move from deployer to provider โ€” with significantly heavier obligations including conformity assessment. Many government IT integrators and SIs are sitting on this boundary without recognising it.

    โ‘ก Japan's procurement baseline is converging toward international standards โ€” faster than its legislation. The controls visible in GENNAI โ€” security review, prompt logging, output handling rules โ€” will become what ministerial staff regard as normal, and what they expect from any private AI tool subsequently approved for use. For vendors selling to Japanese government, this de facto standard is already the relevant one.

    โ‘ข The "deployer" concept is the most exportable idea in the EU AI Act. Even countries with no plans to adopt the EU AI Act will find it conceptually difficult to argue that governments using high-risk AI in employment, welfare, justice, and border decisions should be exempt from accountability. The FRIA will appear in other regulatory frameworks. The logging obligation will follow. Boards and executives should be building toward this standard โ€” not waiting for it to become mandatory in their jurisdiction.

    โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€

    โ–ถ Next (Series #3 ): The Fundamental Rights Impact Assessment (FRIA) โ€” what it actually requires, why most public-sector AI deployments would fail it today, and what this means for government AI procurement in Japan, the UK, and beyond.

    โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€

    This article draws on the AI Governance Policy & Implementation Framework (Enterprise Version 2.0), Nozomu K., AI Governance Lab, Published 14 December 2025. Cite as: Nozomu K., AI Governance Policy & Implementation Framework (Enterprise Version 2.0). This is not legal advice.

    #AIGovernance #EUAIAct #AIDeployer #Article26 #PublicSector #GovTech #AICompliance #FRIA #FundamentalRights #ResponsibleAI #DigitalGovernment #CAIO #GENNAI #JapanAI #AIPolicy #AILeadership #RegulatoryStrategy #GovernmentAI #AIRisk



    For NRT Framework implementation and AI Compliance Audits: ๐Ÿ“ฉ ai.governance.jp@gmail.com (AI Governance Lab)


    ใ“ใฎ่จ˜ไบ‹ใฏnoteใƒžใƒใƒผใซใƒ”ใƒƒใ‚ฏใ‚ขใƒƒใƒ—ใ•ใ‚Œใพใ—ใŸ

    noteใƒžใƒใƒผ ไปŠ้€ฑใฎ็‰น้›†ใ€Œๆ–ฐไฝ“ๅˆถใฎใ‚จใƒŒใƒ“ใƒ‡ใ‚ฃใ‚ขๆฑบ็ฎ—ใ€AIๆ ชใฏๆŒใก็›ดใ™๏ผŸใ€
    ย 
    ย 
    ็‹ฌ็ซ‹็ณปAIใ‚ฌใƒใƒŠใƒณใ‚นใƒปใ‚นใƒˆใƒฉใƒ†ใ‚ธใ‚นใƒˆใ€‚่กŒๆ”ฟ็ตŒ้จ“20ๅนด๏ผˆ็ตŒๆธˆๅฎ˜ๅบใ€ๅœจๅค–ๅ…ฌ้คจใ€ๅ€‹ไบบๆƒ…ๅ ฑไฟ่ญทๅฏฉๆŸปไผšๅง”ๅ“ก๏ผ‰ใ€ๅคงๅญฆใงใฎ็”ŸๆˆAIใ‚ฌใ‚คใƒ‰ใƒฉใ‚คใƒณๅฐŽๅ…ฅใƒป่ช่จผ่ฉ•ไพกๅฎŸ็ธพใชใฉใ‚’ใ‚‚ใจใซใ€็”ŸๆˆAIใฎๅ€ซ็†ใƒปๆณ•ๅ‹™ใ‚’ไธ–็•Œใซ็™บไฟกใ€‚2025ๅนดใซๆ—ฅๆœฌใงๅˆใ‚ใฆ่ฉณ็ดฐใชAIใ‚ฌใƒใƒŠใƒณใ‚นใƒใƒชใ‚ทใƒผใจAIๆดป็”จๆ–น้‡ใฎใƒ†ใƒณใƒ—ใƒฌใƒผใƒˆใ‚’ๅ…ฌ้–‹

    ใ‚ใชใŸใธใฎใŠใ™ใ™ใ‚