Japan AI Guidelines for Business adopt a risk-based approach to AI system categorisation following Hiroshima AI Process principles + conceptually aligned with EU AI Act tiering though voluntary rather than mandatory. (1) Risk Tiering Framework: (a) High-Risk AI - applications with significant impact on fundamental rights + safety + critical decisions (healthcare diagnosis + autonomous vehicles + judicial decisions + recidivism prediction + employment screening + credit scoring + biometric identification + critical infrastructure operation); (b) Limited-Risk AI - chatbots + AI-generated content + recommendation systems with user impact requiring transparency; (c) Minimal-Risk AI - spam filters + game NPCs + general productivity assistance; (d) Prohibited AI Practices - social scoring + manipulative + exploitative + biometric categorisation without consent (aligning with EU AI Act Article 5). (2) Risk Assessment Methodology: (a) Pre-Development Risk Assessment - foreseeable harms identification; (b) Use Case Analysis - intended + reasonably foreseeable misuse; (c) Affected Stakeholder Impact - users + non-users + society; (d) Severity Likelihood Reversibility framework; (e) Per-Deployment Re-Assessment; (f) ISO/IEC 23894 AI Risk Management alignment; (g) NIST AI RMF Function Mapping (Map + Measure + Manage + Govern). (3) Generative AI Specific Categorisation: (a) Foundation Models (Kiban Moderu 基盤モデル) - large-scale pre-trained models including LLMs + multimodal; (b) AISI capability thresholds for advanced models (similar to EU AI Act systemic risk threshold 10^25 FLOPS); (c) Frontier AI - General Purpose AI Models (GPAI) requiring additional scrutiny; (d) Generative AI risks - hallucination + copyright + deepfake + watermarking + manipulation; (e) AISI evaluation for frontier models recommended. (4) Sector-Specific High-Risk: (a) Medical AI - PMDA Pharmaceutical and Medical Devices Agency approval + MHLW guidance; (b) Financial AI - FSA AI Guidelines for Financial Services (September 2024) - high-risk for credit + insurance + AML decisions; (c) Autonomous Vehicles - MLIT regulation + Road Traffic Act amendments; (d) Judicial AI - Ministry of Justice considering use; (e) Educational AI - MEXT guidance; (f) Employment AI - MHLW guidance on hiring AI; (g) Critical Infrastructure - METI critical infrastructure cybersecurity guidelines. (5) AI System Inventory Requirement: (a) maintain comprehensive AI system register; (b) per system include - purpose + risk tier + lifecycle status + stakeholders + data sources + foundation model dependencies + deployment environment + monitoring approach; (c) inventory in 7+ Japanese corporations sample - typical 50-500 AI systems per major enterprise; (d) inventory governance role - AI Officer + DPO + CISO joint oversight. (6) Risk Treatment per Tier: (a) High-Risk - DPIA + Pre-Deployment Approval + Continuous Monitoring + Incident Reporting + Independent Audit; (b) Limited-Risk - Transparency + User Notification + Opt-Out + Documentation; (c) Minimal-Risk - Voluntary Best Practice + Inventory + Aggregate Reporting; (d) Prohibited - Cease + Mitigate Past Use + Compliance Verification. (7) AI System Lifecycle Stages: (a) Conception + Design - requirements + risk assessment + ethics review; (b) Development + Training - data governance + bias mitigation + safety; (c) Validation + Testing - benchmarks + red-team + adversarial robustness; (d) Deployment - phased rollout + human-in-loop + monitoring; (e) Operation + Monitoring - performance + drift + incident + harm tracking; (f) Update + Retraining - re-evaluation + re-approval; (g) Decommission - secure deletion + downstream notification. (8) Risk Communication: (a) Board + Senior Management quarterly reporting; (b) Customer + User affected stakeholder disclosure; (c) Regulator + AISI engagement on high-risk; (d) Industry self-regulation participation (Japan Deep Learning Association + JADMA + Japan Internet Society of Information Communications + Robot Revolution Industrial Cooperation Association). (9) AI Bill Transition: (a) AI Bill currently in Diet expected to mandate certain risk-tier obligations; (b) High-risk + Prohibited categories likely become legally enforceable; (c) Penalties + administrative fines + criminal sanctions for severe violations; (d) AISI evaluation mandate for frontier models; (e) Sector regulators retain authority; (f) Sandbox + Safe Harbor for innovation. Coordinates with EU AI Act + US NIST AI RMF + OECD AI Principles + Hiroshima AI Process Code + AISI Japan + ISO/IEC 23894 AI Risk + ISO/IEC TR 24028 + ISO/IEC 42001 AI Management + FSA AI Guidelines Financial Services + PMDA + MHLW Medical AI + MLIT Autonomous Vehicles + MEXT Educational AI + Critical Infrastructure Cybersecurity + Japan Deep Learning Association + Robot Revolution Industrial Cooperation. Japan AI Guidelines Risk-Based Categorisation applies.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.