tech_data · EU AI Act Article 10 · NIST AI RMF · ISO 42001
Data Poisoning / Supply Chain
Poisoned training data — prove data hash at ingestion or model compromise indefensible
—CMA agent liability guidance (live)
Regulatory exposure — commonly overlooked:
Enterprise fine-tunes on poisoned dataset — enterprise owns breach, data vendor owns ToS. · CMA: you are liable for your AI agent like an employee — applies to every agentic commerce vertical.
TAM / Exposure
Supply chain attacks on ML training · nation-state priority
Insurance lines
Cyber · E&O · Crime
Exhibit authority
UK CMA Agent Liability Pack — Data Integrity + Poisoning Detection Pack
Global leaders
CISA · NIST · Microsoft · Google · OpenAI · CMA · ICO · FCA
SolarWinds parallel
AI supply chain compromise — receipt at data ingestion is detection point.
ISO 42001
AI management system requires data integrity controls — receipt satisfies control.
Incident response
CISA reporting — poisoned model decision receipt proves attack vector.
[United Kingdom] Post-Brexit divergence
UK not bound by EU AI Act but CMA/ICO more aggressive on agents.
[United Kingdom] London market
Lloyd's syndicates need AI decision audit for specialty lines.
[Text / Chat] Modality hook
Baseline — all frameworks apply to text decisions.
7 mandate layers (live)
Regulatory ClockCountdown to operative regulatory deadline — NAIC adoption, GSE mandate, EU transposition.Open →
Domain ClassifierIndustry-specific SAFE/CRISIS/VIOLATION with regulatory framework mapping.Open →
Exhibit / Filing PackRegulator-ready external validation — NAIC Exhibit D, Fannie QC, EU FRIA, FDA Part 11.Open →
Mandate RegistryEnroll deployers/insureds under vertical-specific governance mandate.Open →