How this was made. The version number counts drafts of the text. It does not measure the inquiry behind it, which has run over days and across several AI systems, with argument between those systems and within them, directed, refused and repeatedly redirected by the author. The source material was AI-generated, and then adversarially and iteratively refined across a range of tools — systems built by different companies in different jurisdictions, set against each other and against the author. No one of them produced this text, and no one of them reviewed it alone. The plurality is deliberate rather than incidental. A single model carries a single set of priors about which sources are authoritative, and this series argues that an evidence base narrowed in exactly that way is how a contested question comes to look settled. Using one model to investigate that claim would have been the claim refuting itself. To name a single model on it would credit that model with work that was neither its own nor done in a single pass. The plurality was also necessary, and the record should say why. In drafting, the assisting model repeatedly led with United States institutional sources — a national laboratory, an industry association, a market study nineteen years old — and presented conclusions drawn from them as the state of knowledge. On one occasion European measured data contradicting those conclusions was present in the same research return and was placed below them. Framings were proposed that would have argued against this series’ own position using that evidence base, and offered as rigour. Each was refused by the author and the material rebuilt. That is the mechanism these documents describe, occurring in their own making, and it is recorded because a series arguing that evidence bases narrow without anyone deciding to narrow them cannot credibly claim its own production was exempt. The framing, the corrections and the judgements are the author’s, and so are the errors. How this site is written sets out what is declared on every piece, who checks it, and where the per-piece record lives.
Less than most people assume. The Public Service AI Framework, issued by the Government Chief Digital Officer in January 2025, states in its own words that “Agencies are encouraged to align… however it’s not binding.” The Algorithm Charter of July 2020 is a commitment agencies sign, not a rule they are held to.
The one binding instrument is narrow: the Biometric Processing Privacy Code 2025, which commenced in November 2025 and does real work within its scope. Outside that scope, the position is guidance.
It is an argument that nobody has required them yet, which is a different claim. The piece scores each requirement against what instruments actually say rather than against what would be convenient.
One example makes the point: NIST SP 800-53 control IA-9, service identification and authentication, is written, published and available — and appears in no baseline, not Low, not Moderate, not High. It was not rejected. It was made optional, and optional controls are the ones that do not get bought.
No, and this is the practical finding. Every one of the twelve is a property of an arrangement between an institution and a supplier, and the place that arrangement is written is a procurement schedule.
A government that wants these in force can require them of its own suppliers on Monday. That lever already exists, it needs no legislation, and it moves a market faster than a statute that applies from 2027 — which is when the EU AI Act’s logging obligation at Article 26(6) begins to bite, for high-risk systems only, at six months of retention.
They are written to. The same twelve apply to a district council, a medical practice, a school board and a golf club, and the reason is that none of them describes a scale — each describes a property of an arrangement.
What does not scale is purchasing power. A club secretary can put the questions to a supplier and will generally get no useful answer, because the club is not a customer whose departure would be noticed; a government asking the same questions in a procurement rule gets answers, because they are a condition of a market worth having. Durability is the second difference: a small committee can adopt the twelve on a Tuesday and drop them next year with nobody noticing, which is why the piece argues a requirement lasting a single electoral term is not a requirement.
There is a concrete proposal on the table rather than a wish. In Our Own Hands, offered in June 2026, sets out seven commitments and a three-phase pathway, framed so that it can be signed by parties that disagree about most other things.
Whether it is taken up is not something this series can report. What it can say is that the alternative — an AI policy that changes with each government — produces suppliers who wait rather than build, because the requirement they are being asked to meet has a known expiry.
An organisation holding all twelve that is nonetheless unable to answer for what its systems did, or one holding almost none that consistently can. The twelve are a floor, not a complete regime, and the piece says so: they do not make a system good, and they are not sufficient.
What this publication does not claim, and what is outstanding against it in the register.
Nothing outstanding in the register. Every claim in this publication has its evidence recorded, and no question against it is parked. That is a statement about this publication on the date shown above, generated from the register rather than asserted, and it will change when the register does.
Alongside: the publication · sources and provenance