01What happens.
A regulator’s published documents are the working interpretation of the law for the organisations it oversees. AI now reads those documents to answer compliance questions on their behalf. Where a link inside a document is dead, the claim that rested on it stands unsupported, and the answer is built without the law.
02Why it matters.
If you lead a regulator, your published estate is now read by machines answering for the firms you supervise. A firm asking AI what the rule requires receives an answer assembled from your documents, with the gaps filled from wherever the machine can reach.
A body that holds others to a published standard is seen to be unable to keep its own record in order, which is an embarrassment at best.
The case figure
20% of the links inside the regulator’s legal and regulatory documents were broken, including links to the legal basis of some of them.
AAAnow observed finding, published in the AI Readiness paper, Jun. 2026.
03The case.
A national financial regulator publishes its key regulatory and legal documents as PDFs. The documents contain links to other content: the legislation they implement, related guidance, consultation papers and the pages that explain how the rules apply. That is how a reader, human or machine, follows a rule back to its authority.
The regulator’s documents were read as AI reads them. 20% of the links inside them were broken. Some pointed to pages that had moved, some to documents that had been superseded, and some to material that had been removed. Among the broken links were those to the core legal basis for some of the documents, the legislation the rules were made under.
To a person reading on screen, the documents looked complete. The text was there, the references were there and the links looked like links. The reader found out a link was dead only by clicking it.
To a machine, a dead link is detail out of reach. Whatever the linked page held cannot be read or checked, so an answer that leans on it is built without it. The claim in the document stands, and nothing behind it can be verified.
A compliance officer at a regulated firm, asking AI what a rule required and receiving a confident answer, had no way of seeing that the answer had been assembled without the legislation behind it. The answer read as the regulator’s position, and the officer acted on it.
Regulated firms asking AI what a rule required were receiving answers assembled from documents whose foundations could not be reached. Where the machine filled the gap from other material, older or unofficial, the answer carried the regulator’s authority without the regulator’s content.
The regulator’s own staff, reading the documents on screen, had no reason to suspect the references, because the documents were the versions they had published and the links had worked when the documents were made. Links break as pages move and documents are superseded, and nothing inside a PDF tells its owner that a reference has died. The regulator had no tool reading its documents the way the regulated firms’ systems were reading them.
04Exposure.
The regulator sets the standard for the record-keeping of the firms it supervises, and a published estate in which 20% of links are dead, including links to the law, is a finding the regulator would make against a firm. Beyond the regulator’s standing, a working interpretation of the law is in circulation that the regulator did not write and cannot stand behind.
A firm acting on an unsupported answer carries the compliance cost of getting it wrong, and the regulator carries the correspondence, the clarifications and the corrections that follow, while the firms it supervises carry the cost of finding out. The Open Data Institute found the same unreliability in AI answers built on GOV.UK content, with some answers significantly wrong, including on eligibility (ODI, Feb. 2026).
05What AI Readiness changes.
What the regulator could not see from the inside was which references had died, because the documents looked complete on screen and the links had worked when they were made. AAAnow reads published documents as AI reads them and evidences the links that no longer resolve, the references that point to superseded material and the documents whose legal basis cannot be reached. With that evidence the regulator can repair and republish in an order set by exposure.
Your team makes the changes; AAAnow provides the assessment, the evidence, the guidance and the review.
06Related use cases.
Residents get different answers to the same question, and the cost lands on the council
Contradiction across a public estate and what it costs.
The position of the regulated firm that is held to what it has published.
Funding stopped on submissions that looked fine
Documents that read correctly to a person and differently to a machine, in a public body’s decisions.
07References.
- 20% of links broken, including the core legal basis of some documents: AAAnow observed finding, published in the AI Readiness paper, Jun. 2026.
- Open Data Institute, Feb. 2026. theodi.org/news-and-events/news/new-research-questions-the-trustworthiness-of-ai-chatbots-for-government-informationtheodi.orgWe do not enable external links - please copy and paste, you are then certain as to the link being opened
Next step
Talk it through, in your terms.
The briefing sets this case in your organisation’s terms, and the discussion follows your questions.
Observed, anonymised. No organisation is identified.