The Screen You Never Saw

2026-06-15

Most leadership teams worry about the obvious failure: a public AI system tells a customer something untrue, the customer screenshots it, and the company spends a week cleaning up. That failure is loud, and because it is loud, it is survivable.

The more expensive failure is silent. It happens inside a screen the company never sees.

In 2026, the institutions that decide whether to do business with you have quietly handed the first read to machines. Banks run automated adverse-media checks at onboarding. Private equity and corporate acquirers have moved due diligence onto AI agents that resolve corporate entities, map ownership, and synthesise negative news into source-cited risk profiles. Insurers, procurement teams, and counterparties run versions of the same process. What used to take an analyst days now takes minutes — and increasingly, the analyst is a public AI system asked to assemble everything it can find about your organisation.

This matters because of a problem the compliance industry has lived with for a decade. Adverse-media screening produces false positives at staggering rates — by some estimates up to 90 to 95 per cent of hits are noise. The classic cause is mundane: a name collision, where your company or an executive shares a name with someone genuinely sanctioned, charged, or disgraced. Screening teams have spent years trying to filter that noise out.

Public AI systems introduce a sharper version of the same risk. They do not merely over-flag real events. They can manufacture events that never occurred — a regulatory action that was never brought, a lawsuit that was never filed, a settlement that does not exist — and present it in the confident, tidy prose that reads like a verified finding. We already have proof of how far this goes with individuals. In cases that reached settlement and litigation this year, public AI systems generated fabricated criminal records and invented court documents about a named person, complete with the appearance of sourcing. A system that will invent a criminal record for a person will invent an enforcement action for an organisation. And enforcement actions, lawsuits, and sanctions hits are precisely the categories an adverse-media screen is built to surface.

Now combine the two failures. A screening process already tuned to over-flag, fed by a system capable of inventing the very findings it is hunting for. The output is not just more false positives. It is false positives that look like genuine, cited intelligence.

Here is the part that should keep a board awake. When a public AI system says something wrong to your customer, there is at least a chance you hear about it. When it says something wrong inside a counterparty's compliance workflow, you almost never do. The deal slows without explanation. The account application sits "under further review." The partnership conversation goes quiet. Decline reasons in regulated screening are deliberately vague and legally guarded — no one is going to call to tell you that a machine read something that was not there. You lose, and you never learn why.

This is a different kind of exposure from the reputational risk most communications teams are equipped to handle. There is no article to rebut, no journalist to call, no post to report. The judgement happened in a room you were not in, based on a record you cannot see, and it has already shaped a decision about your access to capital, customers, or coverage.

You cannot correct a verdict you do not know was reached. So the only workable response is to move upstream of the screen. Before a counterparty's system assembles its account of your organisation, you need to know what public AI systems already say when someone queries them about your company, your leadership, and your jurisdiction — across the major systems, not just one, because they do not agree with each other. You need to know which claims are accurate, which are stale, and which were never true to begin with.

Clarity about that — knowing the AI-generated version of your organisation that strangers are acting on — is becoming a precondition for being trusted by the institutions that screen you. The companies that treat it as part of their risk posture, rather than a communications afterthought, will be the ones still getting to the meeting. The rest will keep failing screens they never saw.