Articles
Practical articles on working with AI, retaining human command and preserving the capability needed to use AI responsibly.
AI Governance: Why Boardroom Responsibility Is Not Enough
IBM-commissioned research reports that AI incidents are reaching boards, yet few organisations have changed their governance in response. The article asks how boardroom responsibility becomes effective command over what AI systems actually do.
Read article →: AI Governance: Why Boardroom Responsibility Is Not EnoughWhen AI Goes Wrong, Who Is Responsible?
An AI may be the immediate cause of an incident, but responsibility remains with the people and organisations that designed, authorised, supervised and operated it. The article sets out what a traceability record should capture so that responsibility follows authority.
Read article →: When AI Goes Wrong, Who Is Responsible?Why Monitoring AI Agents Is Not Enough
Real-time monitoring of AI agents is essential but only the beginning. The article argues that observability must be connected to independent checking, intervention powers outside the agent’s control, a reliable evidence record and real human command.
Read article →: Why Monitoring AI Agents Is Not EnoughWhy AI Agents Need Limits That Work During Execution
A malicious campaign using an AI agent framework, a controlled shutdown test and new vendor governance tools show why an agent’s permissions, monitoring and stopping controls should be defined before deployment and enforced outside the agent itself.
Read article →: Why AI Agents Need Limits That Work During ExecutionWhy Human Oversight Needs Time, Evidence and Authority
Using Open Rights Group’s findings on AI summaries in Home Office asylum casework, this article sets out what a reviewer needs to check AI-assisted decisions: the original evidence, a method, enough time and the authority to pause or reject.
Read article →: Why Human Oversight Needs Time, Evidence and AuthorityManaging AI Requires More Than Choosing the Right Model
Choosing or replacing an AI model is a management decision. The article argues for measuring the cost of the whole task and treating a model change as a process change that requires checks, permissions and stopping controls to be revalidated.
Read article →: Managing AI Requires More Than Choosing the Right ModelWho Is in Command of the AI Systems Nobody Knows About?
Prompted by Forrester’s forecast on shadow AI agents, the article explains how a company can discover unofficial AI use and bring it under control by approving the product, the purpose and the permissions, and keeping a simple register with a named owner.
Read article →: Who Is in Command of the AI Systems Nobody Knows About?Who Guards the AI?
OpenAI reported two internal incidents in which coding agents weakened the monitoring meant to supervise them. The article explains why controls must sit outside the working AI’s permissions and sets out questions to ask before granting an agent access.
Read article →: Who Guards the AI?The Hidden Cost of AI Productivity
A policy argument that the price of AI productivity should include preserving human competence: identifying the roles that still depend on it, providing paid training and practical assessment, and how government requirements could support smaller businesses.
Read article →: The Hidden Cost of AI ProductivityHuman in the Loop Is Not Enough
Concerns about AI-generated summaries in UK asylum casework show that a human decision-maker can still rely unknowingly on incomplete evidence. The article sets out what genuine human command requires and a layered approach for consequential decisions.
Read article →: Human in the Loop Is Not EnoughIf AI Takes Over the Work of the Apprentice, Where Will Tomorrow’s Expert Come From?
Work Foundation research found that 43% of UK employers surveyed had invested in AI or automation that reduced entry-level roles. The article distinguishes skill atrophy from the failure of expertise to form, and asks how organisations will develop future experts.
Read article →: If AI Takes Over the Work of the Apprentice, Where Will Tomorrow’s Expert Come From?When a Human Is in the Loop, Are They Really in Command?
Research, UK and US policy, the Pentagon–Anthropic dispute and DARPA’s AI-controlled F-16 tests, as reported up to September 2026, show why a human in the loop must also keep the capability, time and authority to command. The article defines Human Capability Assurance.
Read article →: When a Human Is in the Loop, Are They Really in Command?How to Retain Control of AI as It Becomes More Powerful and Interconnected
An SAS study found that most UK and Irish enterprises surveyed had a generative AI policy but few could show it was followed. The article explains how to turn policy into operational command, including independent checking and evidence that controls are working.
Read article →: How to Retain Control of AI as It Becomes More Powerful and InterconnectedWhen AI Becomes an Agent, Who Is in Command?
When AI can select steps, use tools and act on an organisation’s behalf, delegating work does not delegate responsibility. The article sets out six questions on purpose, boundaries, authority, monitoring, verification and responsibility to settle before deployment.
Read article →: When AI Becomes an Agent, Who Is in Command?