Content Safety
The Sovereign Blacklist: What the NHS FDP Lock-In Teaches Us About Execution Plane Governance. 🟢 Executive Summary When Freedom of Information (FOI) data reveals that 29 of 137 NHS hospital trusts listed as "realising benefits" from the £330m Federated Data Platform (FDP) haven't launched or deployed
Agentic Behavioural Architecture
🟢 Overview Banks are now detecting third‑party AI agents inside customer accounts. Some agents act without user instruction. Executives can’t reliably tell human activity from autonomous behaviour. This isn’t fraud — it’s agentic execution inside sovereign financial systems. 🟢 The core failure is credential abdication. Customers are giving plaintext
EU AI Act
🟢 Why the Mills Review matters FCA: “The Mills Review examines how autonomous artificial intelligence could increasingly reshape retail financial services and regulatory oversight.” * Delegated AI decision-making now part of sector operations. * Frontier-AI cyber capabilities + model concentration = system-wide risk. FCA signals AI is no longer peripheral. 🟢 Consumer Duty
EU AI Act
🟢 Main Issues The Commission’s investigation is not about isolated scams; it’s about systemic governance failures in a VLOP (Very Large Online Platform). * Fraudulent listings reveal gaps in risk identification, mitigation and marketplace integrity. * Phishing pathways emerge from data-flow design and messaging architecture, not user error. * Transparency failures
ACW
🟢 Overview * Agentic AI shifts risk from “what the model can do” to “how the system behaves in deployment.” * Behavioural failures emerge from interactions between agents, humans, governance, workflows and organisational context. * Identical models can produce radically different risks depending on access, authority and trust. 🟢 Behavioural Failure Modes (Turing taxonomy) * Goal
ACW
🟢 Disclosure ≠Legitimacy * Organisations treat AI disclosure as a compliance box, not a trust‑design problem. * Research shows disclosure reduces trust; hiding AI use and being caught reduces it even more. * The mechanism is legitimacy: people expect human effort in tasks like writing, analysis, and decision support. * Without legitimacy, even high‑
ACW
🟢 Misunderstood Control? HITL may be the Most Misunderstood Control in AI Governance. * Most HITL in industry is theatre — robotic checkpoints added for policy optics, not real control. * Organisations drift toward “faster/cheaper” laziness; outsourcing exposed the pattern: theatre to win contracts → theatre to avoid SLA penalties. * Real HITL ≠a human