Case Studies and Anti-Patterns

The earlier chapters argued from principles and from the two fictional organisations. This chapter checks the argument against public cases, in which real organisations, inquiries, courts and regulators have recorded what went wrong and sometimes what went right. It concerns all five leadership functions: L1 Orient, because several cases turn on what the leaders in charge were told and by whom; L2 Decide, because they turn on who took a decision and on what evidence; L3 Organise, because the recurring fault is a structure in which nobody owned the whole; L4 Enable, because a dashboard or a target replaces the work of developing and checking people; and L5 Answer, because every case ends with a question of who answered, to whom, and whether they could.

The method is strict, and the reader should know it before the cases. Every case below is described only from a primary or authoritative record that was opened and read in preparing this chapter: a parliamentary inquiry report, a data protection authority's investigation report, a regulator's order, a supervisory decision, a head of government's own words, or a company's own post-incident report or press release. Where the record is a company's statement of its own results, it is described as a claim. A case that could not be verified that way is not analysed, and the section on method lists them. Interpretation is the book's own and is labelled as such. The cases are observable in the sense of AI Governance, Chapter 22: they happened and are documented (221, Nordqvist, 2026).

The chapter takes five conclusions from the earlier volumes as the lenses through which it reads the cases. The July 2026 OpenAI and Hugging Face incident failed through correlated controls, among them an alert that reached people who did not hold the stop decision (AI Governance, Chapter 19, F28) (221, Nordqvist, 2026). Agents acting beyond their mandate are observable, and the lasting response has been to move controls from the agent's compliance into the environment (AI Threat, Chapters 25 and 35, F67) (224, Nordqvist, 2026). In the first exercise on the defenders' own AI, the action control held while the case store accepted unlabelled text, and each defect needed a re-test (AI Blue, Chapter 15, F114) (227, Nordqvist, 2026). The loop fails in recognisable ways, each a step that looked reasonable, and the controls are few: independence, evidence age, a stop path outside the system and a testing right (AI Purple, Chapter 26, F140) (228, Nordqvist, 2026). The first honest numbers of a verification programme are lower than the first dashboards (AI Purple, Chapter 13, F141) (228, Nordqvist, 2026). From this book, Chapter 4 supplies the three tests of the answerability principle, Name, Explain and Reach, the delegation ladder with the four Level 4 prohibitions, and the rule that a claim carries its method and its date.

Primary topics explored

1. What this chapter builds on
2. How the cases were chosen and what was dropped
3. Cases, by mechanism
4. What recurs
5. Anti-patterns
6. Lessons by function