-By Dr Joanna Michalska Over the past several weeks, this series has traced a consistent gap. The gap between what organisations know and what they can do. Between signals that appear and decisions that follow. Between the oversight roles that exist and the intervention authority that is exercised. This time, the same gap appears in a different form. One that is growing more visible as AI governance investment accelerates. The gap between observing a system and being able to

-By Dr Joanna Michalska The last article’s argument was about escalation as a design problem. About the structural conditions that prevent concerns from travelling upward, and the credible intervention deficit that keeps people silent even when they know something is wrong. This week, the question moves to the moment after escalation. Assume the concern has reached the authority. A decision now has to be made. The system has not clearly failed. No threshold has been formally

-By Dr Joanna Michalska When an AI system produces outputs that worry the people closest to it, but no threshold has formally been breached, who decides whether that is a problem worth acting on? This is where a quiet but consistent governance failure appears. Not the absence of escalation pathways, but the absence of defined intervention thresholds. The Borderline Problem AI systems do not typically fail in obvious ways; they drift. They produce outputs that are statistical

-By Simon Shobrook Governance as a procurement gate The UK has no single AI law. The government has deliberately taken a sector-regulator approach, asking the ICO, FCA, CMA and others to apply existing frameworks to AI in their respective domains. But that doesn't mean procurement teams have no obligations. Cabinet Office guidance, the Algorithmic Transparency Recording Standard, and growing pressure from central government on audit-readiness are already shaping what gets

-By Simon Shobrook There’s a lot of talk about concepts, policies, and frameworks. But very little is actually deployed and proven in practice. The reality? The horse has already bolted. By autumn 2023 (ancient history in the AI universe!), the National Audit Office had identified 74 AI use cases already deployed across surveyed government bodies. That’s a conservative baseline. Since then, Cabinet Office transparency publications under the Algorithmic Transparency Record

-By Simon Shobrook Discussions of AI performance typically begin with accuracy. A model is described as “95% accurate” or as outperforming a benchmark. These statements are useful. They provide a high-level indication of model capability and allow for comparison across approaches. However, once an AI system is deployed into an operational environment, accuracy is no longer the primary determinant of outcomes. Consider pre-AI rule-based systems, such as highly accurate f

-By Simon Shobrook Governance frameworks often fail simply because they aren’t put into practice. Add AI into the mix, and things can quickly spiral out of control. Policies get written. Principles have been agreed upon. Committees get formed. And then reality happens. AI systems go live. Models change. Risks evolve. Suppliers update without visibility. And the “framework” sits in a document somewhere, disconnected from what’s actually happening. That’s the gap. If governa

-By Dr Joanna Michalska In recent weeks, I have examined how automation redistributes authority and how many organisations are not structured to act at the speed of their own systems. This week turns to whether governance that satisfies audit can actually deliver intervention when it matters. AI governance has entered a new phase. The European Union Artificial Intelligence Act is moving from principle toward enforcement, and U.S. states are expanding oversight capacity. Pu

-By 🔐Rajeev Chakraborty Imagine four experts in a room. One from Brussels, one from Washington, one from London, one from a tech company. You ask them: What does "fairness" mean when an AI makes decisions about people? All four nod confidently. All four give completely different answers. This is AI ethics today. Not disagreement about whether fairness matters. Everyone agrees it matters. The problem is that nobody uses the same words to describe it, measure it, or govern i

-By 🔐Rajeev Chakraborty If Horizon taught us anything, it’s this: when complex systems are shielded from scrutiny, the consequences fall on people. AI only raises the stakes. Picture this: You're a subpostmaster in a small British village. You've run the local Post Office for fifteen years. Your neighbours trust you. You know their children's names. Then one Tuesday, the computer system shows you've lost £40,000. You haven't. But the screen says you have. And the organisa

-By Simon Shobrook Picture a procurement meeting. A vendor is presenting their AI system. It screens applications, prioritises cases, and flags risks. The slides are polished. The demo is smooth. Someone asks: "Is this system responsible AI-compliant?" The vendor says yes. Everyone nods. The meeting moves on. But here's the problem. Nobody in that room agrees on what the question actually meant. And nobody agrees on what "yes" actually proved. This is where most AI procure

-By Simon Shobrook Every vendor will tell you their AI system has "full visibility". They'll show you dashboards. They'll mention logging. They'll point to audit trails. And technically, they're not lying. The system does record things. Data goes in, decisions come out, and somewhere in between, a log file captures what happened. But here's the question nobody asks: can you actually see what's going on? Because logging and visibility are not the same thing. And confusing th

- By Rajeev Chakraborty Your organisation probably has an AI ethics framework by now. Maybe you adopted one from government guidance. Maybe a working group wrote one. Maybe you signed up for industry principles. They all say roughly the same thing: be fair, be transparent, be accountable. The problem isn't the principles. The principles are universal, well-intentioned, and almost entirely useless when you're trying to govern an AI system in production. You deploy an AI system

Remember when GDPR hit and everyone panicked? Well, here we go again. The EU AI Act is live, and honestly? Most companies I talk to are flying blind. Here's the thing - while your competitors are scrambling to figure out what "high-risk AI systems" even means, there's a massive opportunity sitting right in front of you. I've been working with companies across Europe and the UK, and the ones getting this right aren't just checking compliance boxes. They're actually turning AI

Did you know that: Under (Chapter II, Art. 5) EU AI Act, which most AI service providers comply with, are prohibited from: - "social scoring, i.e., evaluating or classifying individuals or groups based on social behaviour or personal traits, causing detrimental or unfavourable treatment of those people". Computer Weekly article "DWP 'fairness analysis' revealed bias in AI fraud detection system" which states: "the assessment showed there is a 'statistically significant refer

It’s easy to say “we want ethical AI.” It’s harder to operationalise it. Looking at assurance checklists, a few themes always stand out: • User access management: Who can touch training data, and how often are permissions reviewed? • Vulnerability management cadence: Are model security checks monthly, quarterly, or ad-hoc? • Explainability logging: Is every model decision traceable back to inputs and assumptions? • Sustainability: Do we measure compute costs and energy us

The technology has already woven itself into our lives. The only question is which companies survive the journey. Is AI in a bubble? Probably. Will it crash? Quite possibly. Does any of that matter for AI's long-term trajectory? Not really. Even Sam Altman, the CEO of OpenAI, has acknowledged that investors are "overexcited" about AI. The Bank of England has warned about the risks of a global market correction. Apollo Global Management's chief economist has stated that the cu