A person of standing opens the Financial Times and reads about someone they know. Not a deal they led together. Not a fund they share. A separate matter entirely, one they had no part in and no knowledge of. But the piece names a vehicle, a board, or a venture the two of them both sit in, and from the second paragraph onward the two names travel together. By the time they have finished their coffee, a search for their own name returns the same article on the first page.
Nothing they did caused this. That is the point.
For anyone of standing, whether that is a family office, an ultra-high-net-worth individual, a chief executive, a founder, a board director, or someone in public life, the hardest reputational exposures are rarely the ones that originate with them. They arrive through association. A co-investor, a fellow director, a fund manager, a foundation trustee, a campaign donor, a business partner taken on years earlier and barely thought of since. Each of them carries a profile of their own, and the moment the relationship is documented anywhere a journalist can find it, that profile becomes part of theirs.
The media frame is not “guilty.” The frame is “linked to.” It does not require an allegation against them, a finding, or even a suggestion of involvement. Proximity is the story. Once two names appear in the same paragraph often enough, a search engine treats them as related, an AI model summarises them as connected, and a reader draws the inference no editor ever printed.
What diligence is actually protecting against
The instinct is to vet for wrongdoing, the criminal record, the regulatory sanction, the obvious red flag. Those matter, but they are the easy cases, and they are rarely what causes the damage. The damage comes from the partner who was clean at the point of entry and became a problem afterwards. The fund that was reputable when the commitment was made and attracted scrutiny three years later. The director who joined a board for entirely sound reasons and was named, fairly or not, when that board’s other business came under question. The co-founder, the early backer, the political ally whose later conduct rewrites the story of everyone once photographed alongside them.
The pattern is not hypothetical. When Theranos collapsed, its board carried some of the most respected names in American public life, former cabinet secretaries, a decorated general, a former senator. None was charged with the fraud. None was accused of running it. Yet “former Theranos director” now sits in the public record of each of them, surfaced by a search engine and recited by an AI model whenever their name is checked, attached to a company whose founder was convicted. They lent their standing to a board for reasons that were entirely defensible at the time, and the association outlived the reasoning. The same mechanism has played out closer to home. David Cameron, never found to have done anything unlawful, became so closely tied to the Greensill affair through an advisory relationship that the two names are now difficult to separate in any search of either.
This reframes the discipline. Diligence is not a gate you pass through once. It is a standing condition. The counterparty you checked in 2023 is not the counterparty you are exposed to in 2026, and the only way to know the difference is to be watching. Most people are not. They learn that an association has turned from a journalist, a regulator, or a search result, which is to say they learn too late to shape it.
The cost of being caught in the crossfire is not usually a legal one. It is the unwelcome attention, the file a private bank opens, the question raised in a board meeting, the donor who quietly steps back, the paragraph that now sits permanently on the first page of search. It is being convicted by association in the only court that does not require evidence, the court of the casual reader who searches a name and reads the first three results.
The tool that is supposed to help
Here is the turn, and it is an uncomfortable one.
The response to all of this, increasingly, is to run counterparty checks through AI. An adviser, a private bank analyst, a journalist, a prospective partner’s own team, types a name into a frontier model and reads back what it returns. It is fast, it is cheap, and it feels authoritative. It is also, frequently, wrong.
AI models do not apply recency bias the way a search engine’s first page does. A search result ages. A story falls down the rankings as newer material displaces it. An AI summary does no such thing. An allegation that was withdrawn, a case that resolved in the subject’s favour, a matter that settled without finding, a story that was later corrected, all of these can surface in an AI-generated profile as though they were current and unqualified, stripped of the outcome that resolved them. The model reports the accusation and omits the acquittal, because the accusation was written about more.
The person running the check has no way of knowing this. They see a confident, fluent summary and take it as the state of play. They are reading a snapshot that may be years out of date, decontextualised, or simply wrong, presented in the register of fact.
And the error runs in both directions. The same unreliability that taints a clean counterparty can clear a compromised one, surfacing nothing because nothing has been indexed in the form the model favours. So the exposure runs two ways. A person may be wrongly tainted in someone else’s AI check, an association they cannot see and were never asked about. And they may be relying, in their own diligence, on a tool that is quietly missing the thing they most need to find.
What this means in practice
Three things follow, and they hold whether you are running a family office, leading a listed company, building a business, or serving in public life.
The first is that diligence on associations must be continuous, not transactional. The exposure does not end when the commitment is signed, the appointment is announced, or the partnership is struck. It begins there.
The second is that your own AI profile is now a diligence input for everyone you deal with, and it is one you have almost certainly never audited. What a model says about you is being read by counterparties, banks, journalists, selection committees, and prospective partners as part of their own checks. If it is wrong, you are being mischecked without ever knowing.
The third is that accuracy in the AI layer is not self-correcting. A withdrawn allegation does not withdraw itself from a model. A favourable outcome does not insert itself into a summary. Left alone, the snapshot persists. It has to be actively managed, in the same way the search layer has had to be managed for two decades.
The company you keep has always shaped how you are seen. What has changed is who is doing the looking, how quickly they reach a view, and how rarely that view is checked against the present.
What we see at Pavesen
Two patterns reach us, and the balance between them is shifting.
The first is the client who comes after the fact. The association has already surfaced, the paragraph is already on the first page, and the work is remedial: understanding what is there, where it sits, and what can be done to reduce its prominence over time. This has always been part of the work and it remains so.
The second is now the more frequent request, and it is the more useful one. Clients increasingly ask us to perform the diligence before the relationship is formed, on a prospective partner, a co-investor, an incoming director, a counterparty. Not the surface check a financial professional already runs, the registry search, the sanctions list, the credit file, but the layer beneath it. The archived version of a website that has since been quietly edited. The forum history. The Reddit thread. The local-language press that an English search never surfaces. The social media account that says more than any filing. The story that was published, then unpublished, and survives only in a cache. We assemble these into a single profile of the person as they actually appear across the open record, with the context and the outcome attached, so the client is reading the present rather than a fragment of the past.
The distinction matters, because it is the same distinction the rest of this piece has drawn. An automated check returns a snapshot, fast, fluent, and frequently stripped of the resolution that would change its meaning. Diligence done properly returns the opposite: the full picture, sourced, dated, and read by someone whose job is to notice what the machine omits. One tells you what was written about most. The other tells you what is true now.