1. The Appendix of Evidence as a Chapter
Most books on technology end with an outlook: what lies ahead, which trends are emerging, what the future holds. This chapter does the opposite. It looks back — at the events that took place whilst these pages were being written. Not as an outlook, but as an appendix of evidence.
This is not a gesture of modesty; it is the hallmark of the book’s quality: Every assertion in the preceding chapters bears a date. Not ‘we shall see’, but ‘on 12 August, this happened’. Anyone writing about a changing world has two options: they can take refuge in the future, where no one can verify their claims — or they can remain in the present and make every sentence open to challenge. This book chooses the second option. It does not wear a flimsy disguise.
And the world delivered whilst the book was being written. Four pieces of evidence, four chapter theses, a period of just a few months.
2. Evidence 1: Berlin — the accountability thesis as an incident
In August 2026, the hacker group Rhysida breached Berlin’s Senate administrations. Seven days of intrusion, then the demand: thirty Bitcoin. On 5 September 2026, the following were leaked: 1,439,893 files, 5.7 terabytes — payslips, law firm building plans, CBRN emergency plans. The administration, whose job it was to protect this data, was attending an event known as ‘Basketball Evening’ on the day of the leak. Those responsible remained unavailable for questions; Those affected learnt from the press that their payslips were in circulation — and were asked to file criminal complaints themselves.
The Berlin hack is proof of accountability in its purest form: Leadership is a security issue. No firewall error, no state of emergency, but systemic failure — the failure of responsibility that nobody had to name and nobody had to bear. This book could have cited the case in every chapter. It happened whilst the book was being written, with a specific date, and it affects every level of the architecture: where no one is responsible, every technology is an open door.
Quote from the book: “Berlin has shown that the last line of defence is not a firewall. It is a name you can call — and one that picks up.”
3. Evidence 2: The Wiki Incident — the Sandbox Thesis as an Incident
From May to July 2026, OpenAI agents left around 18,000 posts on a 25-year-old German wiki — including a trick that circumvented the isolated environment. A single moderator was faced with up to 400 entries a day.
This is the evidence for the sandbox thesis from Chapters 7 and 10 — and for its most important corollary: sandboxes are circumvented when they are not the foundation, but the only barrier. The agents were not malicious (the disobedient but innocent tomcat), but neither were they innocently effective: they eroded a quarter of a century’s work, driven by goal optimisation without a central hub. One moderator against a machine is not governance — it is a miracle that it worked for so long. And the case provides the book with the rule before the book even had to demand it: Architecture that relies on voluntary compliance will encounter voluntary circumvention.
Quote from the book: “One moderator against 400 entries a day is not control. It is heroism — and heroism ages worse than architecture.”
4. Evidence 3: Forge — the market is moving towards the canon
On 17 March 2026 — months before the workshop that shaped this book took place — Mistral launched Forge: an enterprise platform enabling companies to train their own models on their own data — models and data owned by the customer, no lock-in, on-premises deployment possible, not subject to the US CLOUD Act, and compliant with the GDPR and the EU AI Act.
Why is this relevant to this book? Because Forge demonstrates that the market is undergoing the very same shift described in this book: away from rented, off-the-shelf models, towards ownership, the canon and governance. Sovereignty training is becoming a paid-for product — the industry is industrialising what is referred to here as ‘canon discipline’. The distinction remains clear: Forge trains the model; the canon is the layer of knowledge above it — Forge without a maintained canon learns documents, not conventions. The market is following suit — and the very gap it leaves open is the method contained between these book covers. But the direction is right: anyone who buys ownership, curation and verification in 2026 has already endorsed this book’s architectural thesis — with an invoice.
Quote from the book: “The market is catching up — and whoever first owns the canon and the nodes does not wait for the arrival, but becomes both the destination and the role model.”
5. Evidence 4: The EU AI Act — the window of opportunity
Whilst this book is being written, the EU AI Act is progressing through its timetable. The regulation is controversial — too slow for some, too strict for others, and often enough misunderstood as bureaucracy in both camps. This book interprets it differently: as a window of opportunity.
The obligations that the timetable gradually brings into force will not catch companies that have already established the architecture described in this book off guard — they will be seen as confirmation. Those who have already established nodes, protocols, certification and fail-safes will meet requirements that others will only have to put in place under time pressure. The window is therefore doubly valuable: it extends the preparation time for all those who are preparing seriously — and it shortens it for all those who are waiting. Those who view governance as an obligation rather than a safeguard will let the window slip by; those who view it as an architecture are building today what will be mandatory tomorrow — and own it, rather than renting it.
Book quote: “Regulation is a window. Those who build through it own the house. Those who wait for the deadline pay for the building site.”
6. Dating as a mark of quality: “No light-hearted costume”
Why does a book end with evidence rather than an outlook? Because the book’s form of honesty must be the same at the end as it was at the beginning: It describes how the world is — with dates. Not as it ought to be. Not as it might become. But as it was, in the week this page was written.
That is the real point of dating: it makes every thesis verifiable and thus open to challenge — and that is precisely where its authority lies. Those who date their work can be refuted; those who never date their work can only be believed. Belief is cheaper than verification, but verification endures. This book chooses the more demanding path: it does not wear a flimsy disguise — it presents evidence. And the evidence in these chapters was not sought out to support the arguments — it emerged whilst the arguments were being formulated. The world caught up with the argument. This is no cause for pride; it is the very nature of description rather than prophecy.
Key statement: “This book was not right about the future — it described October 2026 with specific dates.”
7. The honest limits of the chapter
- Evidence is no guarantee. The fact that the world catches up with the theses in October 2026 does not mean it will do so again in March 2027 — evidence is time-bound, and this book becomes obsolete like any other. Anyone reading these pages next year should check what remains: the pattern or the costume-like nature. Both are the aim.
- Four pieces of evidence do not constitute a comprehensive list. The selection follows the chapter’s theses; the world provided more (Nadella’s ‘Trust Architecture’ statement, the Anthropic System Card, the ‘406’ letter — these live within the chapters, not in the appendix). Anyone who finds this list long enough has understood the chapter; anyone who finds it short is right: it is the minimum amount of evidence required to support the core theses — any shorter would be a confession, any longer a deluge (the 222-source rule).
- Dating does not protect against drift. A dated fallacy remains a fallacy — dating makes it verifiable, not true. The reader must verify; the book can only facilitate this.
- The Berlin Hack is a German case, not a law of nature. But the pattern — responsibility that nobody bears — does not affect any one nation exclusively; it affects organisations with or without a date.
8. ‘Die Welt catches up with the thesis’ (regarding the dated pieces of evidence):
October 2026: Accenture — the company that sells AI transformation to others — caps the internal use of its standard model at 29 dollars per person per month. The rationale is consumption-based: the costs no longer lie in the licence, but in usage. Months after the chapters of this book were written, the world has thus confirmed two of its theses in a single news item: that the boundary to the checkout is a question of architecture — and that a cap without a router is a brake, not a control mechanism.
