Generative AI and Foundation Models require specific governance attention beyond traditional AI risk frameworks per Japan AI Guidelines for Business + augmented by AISI Japan AI Safety Institute + Hiroshima AI Process Code of Conduct + emerging AI Bill. (1) Generative AI Definition: (a) AI systems capable of generating novel content (text + image + audio + video + code + 3D + multimodal); (b) Foundation Models (Kiban Moderu 基盤モデル) - large-scale pre-trained models adaptable to many downstream tasks (GPT-4 + Claude + Gemini + Llama + Mistral + Japanese models like Stockmark + ELYZA + Sakana AI + PFN); (c) General Purpose AI (GPAI) per EU AI Act Article 3; (d) Frontier AI - capability frontier models warranting enhanced scrutiny. (2) Generative AI Specific Risks: (a) Hallucination - confidently stated falsehoods; (b) Copyright Infringement - training data copyright + output reproducing copyrighted content; (c) Deepfake + Manipulation - face/voice swap + disinformation + non-consensual intimate imagery; (d) Privacy - personal information memorisation + extraction attacks; (e) Bias + Stereotype + Harmful Content - amplification + reinforcement; (f) Prompt Injection - direct + indirect (websites + tools); (g) Jailbreak - safety guardrail bypass; (h) Model Extraction + Theft - reverse engineering; (i) Data Poisoning - training data corruption; (j) Misuse for cyberattacks + bio/chem weapons + autonomous replication (frontier capability risks). (3) Hallucination Mitigation: (a) Retrieval-Augmented Generation (RAG) with verified knowledge bases; (b) Citation + source attribution; (c) Confidence scoring + uncertainty quantification; (d) Fact-checking integration; (e) Domain-specific fine-tuning with curated data; (f) Chain-of-Thought + verification; (g) Human review for high-stakes outputs; (h) Hallucination detection benchmarks (TruthfulQA + Japanese benchmarks). (4) Copyright Compliance: (a) Copyright Act 2018 Article 30-4 text data mining exception; (b) opt-out signals from rightsholders (robots.txt + ai.txt + License headers); (c) output filtering to prevent verbatim reproduction; (d) attribution where required (CC-BY); (e) Japan Government Cabinet Office Working Group on Copyright + AI (2023-2024) guidance; (f) US Music Publishers v. Anthropic + NYT v. OpenAI litigation considerations; (g) Japanese rightsholder consortium engagement (JASRAC + Manga Publishers + News Publishers); (h) Generative AI dataset transparency. (5) Watermarking + Content Provenance: (a) C2PA Coalition for Content Provenance and Authenticity Content Credentials standard; (b) SynthID (Google) invisible watermark for images + text + video; (c) Adobe Content Credentials; (d) Microsoft Provenance Service; (e) Meta AI labelling; (f) OpenAI watermarking research; (g) EU AI Act Article 50 mandates watermarking; (h) China Generative AI Measures Article 12 mandates watermarking; (i) Japan voluntary moving toward standardisation. (6) Prompt Injection + Jailbreak Defence: (a) Input sanitisation + filtering; (b) Output filtering + content classification; (c) System prompt hardening; (d) Anthropic Constitutional AI techniques; (e) OpenAI safety + alignment training; (f) Tool use sandboxing; (g) Red-Team continuous; (h) Bug bounty programs; (i) AISI evaluation of guardrails. (7) Frontier AI + AISI Pre-Deployment Evaluation: (a) capability thresholds (compute + parameters + benchmarks); (b) AISI evaluation methodology - capability + safety + alignment; (c) dangerous capability assessment - biosecurity + cybersecurity + autonomous + persuasion; (d) Responsible Scaling Policies (RSP) - Anthropic + OpenAI + Google DeepMind; (e) Pause-and-evaluate protocols; (f) Hiroshima Code of Conduct frontier developer commitments; (g) AISI Network joint evaluations. (8) Generative AI Deployment Safety: (a) Content moderation pipeline; (b) Rate limiting + abuse prevention; (c) User reporting + flagging mechanism; (d) Age verification for adult content; (e) Mental health + crisis intervention triggers; (f) Misuse pattern detection + response; (g) Coordinated Vulnerability Disclosure (CVD); (h) Sector-specific safeguards. (9) Japan-Specific Generative AI Issues: (a) Japanese language + cultural sensitivity in foundation models; (b) Domestic foundation model development (Stockmark + ELYZA + Sakana AI + PFN + Tooku); (c) Cross-lingual capability gaps; (d) Anime + Manga character likeness IP concerns; (e) Japanese voice cloning concerns; (f) Senior + Vulnerable population AI literacy gap. (10) Generative AI in Sector-Specific Applications: (a) Healthcare - medical advice + diagnosis assistance + clinical documentation; (b) Financial - investment advice + AML + fraud detection + customer service; (c) Legal - contract analysis + research + drafting; (d) Education - tutoring + content generation + assessment; (e) Public Sector - citizen service + welfare; (f) Marketing + Advertising - content generation + personalisation. Coordinates with EU AI Act Article 50 watermarking + Article 51-55 GPAI obligations + China Generative AI Measures Article 12 + Hiroshima AI Process Code of Conduct + AISI Japan + AISI Network + C2PA + Content Authenticity Initiative + Anthropic RSP + OpenAI Preparedness Framework + Google DeepMind Frontier Safety Framework + Microsoft Responsible AI Standard + NIST AI RMF + ISO/IEC 23894 + ISO/IEC TR 24028 + Japanese Cabinet Office Working Group on Copyright + AI + JASRAC + Manga Publishers + News Publishers consortium + APPI + PIPC + Japan Deep Learning Association. Japan AI Guidelines Generative AI + Foundation Models applies.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.