What if the AI reading your crisis statement is no longer on your side of the table?

By Tim Sutton, senior reputational adviser · 24 June 2026

AI drafts your crisis statement in seconds. It is also the first to read it, on every desk that matters. So who, exactly, are you writing for?

When the regulator, the journalist and the claimant all read your crisis statement through an AI before a single human forms a view, who exactly are you writing this statement for?

It is the question I now put to anyone dealing with a crisis. And almost none of them have an answer.

Everyone tells you artificial intelligence will speed up your crisis response. Faster drafting, faster monitoring, a first holding statement in ninety seconds instead of ninety minutes. I have heard it from boards, from agency principals, and from earnest panels where nobody in the room had yet been on the firing line. All true, and all beside the point, because speed was rarely the thing that lost a serious crisis.

The room that taught me this

Let me take you back to a room, because the argument is really about that room. A family business I represented, a multi-billion-dollar cross-border liquidation claim that lived for months in legal filings: thousands of pages of pleadings and affidavits, a case that looked indefensible to anyone who only read the paper.

The instinct in the room, the one that very nearly won, was to fight it there. Answer every allegation, correct every error, out-document the documents. I remember the night that instinct lost, somewhere past midnight, when it became plain we could win every line on the page and still lose the only verdict that mattered, the one being formed by people who would never read the page at all.

So we did the opposite. We moved it out of the filings and into human terms, with a single exclusive in the Financial Times, and the story changed shape because the reader changed. A position that had looked indefensible became defensible. The hard part, I realised, was never the drafting (we did that by hand, on a fax machine, remember those?). It was the judgement about who is reading, and what they do with what they read.

That is the part everyone now overlooks. The machines read faster than any of us; they summarise ten thousand documents before the general counsel has finished their coffee. Quite so. But what artificial intelligence has changed is not the speed of your side of the table. It is the identity of the reader on the other side. None of this is universal yet. Plenty of people still read your words directly. But in a serious crisis you plan for where the traffic is going, not where it has been.

Four readers you did not write for

The regulator reads you through an AI model first. Not as policy, not yet, but increasingly in practice. A supervisory team facing a tide of disclosures can run them through tools that flag the one clause that does not reconcile, that cross-reference your statement today against your statement in 2019.

The Financial Conduct Authority has been admirably plain: for now it will write no new AI-specific rules, leaning instead on the frameworks it already has, the Consumer Duty and the Senior Managers and Certification Regime. Which means accountability for what a model surfaces sits exactly where it always sat, with named senior managers and, behind them, the board. The AI does not dilute personal accountability. It sharpens it, because it makes the inconsistency easier to find and harder to explain away.

The journalist reads you through a model first. The exclusive I won in that family matter was won human to human, a relationship, a story told straight. That route still exists, and it matters more than people think. But the first pass on your statement is now often an AI prompt. Summarise this. What is the company not saying? Where is the gap between the words and the filings? The journalist who once skimmed your three hundred words now has a tool that extracts the load-bearing sentence and discards the comfort wrapped around it. The wrapping was the part we used to labour over.

The employee reads you through an AI model first. Your internal note, drafted with such care to reassure, is pasted into a chatbot within the hour. Decode this for me. Should I be worried? The reassurance you wrote for a frightened human is flattened by a tool that has no interest in being reassured, only in the proposition underneath.

And the claimant reads you through the AI model first. This is the one that should keep a general counsel awake. A line written to sound humane, "we deeply regret the distress this has caused", is, to a model trained to find admissions, an admission. Distress. Caused. By us. The Compensation Act 2006 says an apology is not, of itself, an admission of liability. The AI model may not have read the Compensation Act. The claimant's adviser no longer needs to spot the phrase. The AI spots it, flags it, and presents it as an opportunity. The kind word you wrote to a human becomes a sentence in someone else's particulars of claim.

Do you see what has happened to the table? AI did not take a seat on your side of it. It moved into the middle. Each counterparty now has a reader you did not write for, sitting between your words and their decision.

Two objections worth taking seriously

Two sensible objections, and I want to give each its due.

The first, which the trade press has circled all year: can you trust AI in a crisis? PRWeek more or less ran the question. The answers are sensible enough. The models hallucinate, they miss nuance, they produce confident nonsense in a clean font, so keep a human in the loop and check everything. Good advice, and I would give it to a board myself. But notice what it governs. A human in your loop governs your use of AI. It does nothing whatever about the AI in everybody else's loop. You can place the most experienced general counsel in between your statement and the outside world, and they still cannot supervise the model the claimant's solicitor runs the moment your words land. The discipline we have all agreed on guards only what we can control on our side: it cannot control the AI on the other side of the table.

The second, and the most current. Isn't this just the new fashion for treating AI as your audience, the thing everyone now calls AI optimising? No, and the difference is the whole point. Courting AI as an audience, so the model quotes you, ranks you, recommends you, is a real discipline and a useful one. But in a crisis, the AI is not only an audience you are courting. It is a counterparty reading you for the one line it can use against you. You do not get to optimise for it. You only get to be careful in front of it.

There is a sharper edge to this, which is very recent. The most advanced AI models have stopped being available to everyone at the same time. In June the United States issued an export-control order barring foreign nationals from Anthropic's most advanced models, Fable 5 and Mythos 5, and Anthropic switched them off for those users within hours (Reuters). The European Commission called it discriminatory. Whatever you make of it, the effect on this argument is plain: the counterparty reading your crisis statement may be running a model you are now barred from running yourself. It is one thing to be read by a machine. It is another to be outgunned by it, because the sharper machine sits on the other side of a border. The position may ease, Anthropic says it is working to restore access. But I would plan for the version where it does not.

The new craft: one set of words, two readers

Which brings me to the practical core, and it is simpler than the conference circuit makes it sound. You are now writing two documents at once, in the same words. One is for the human, and has to land with warmth and proportion and a recognisable voice, because a board that sounds like a machine in its worst week loses the room. The other is the skeleton the AI extracts once every comforting clause has been stripped away, and that skeleton is what reaches the journalist, the regulator and the claimant first.

The new craft is one set of words that does both: that comfort the human without conceding to the machine. It is harder than it sounds, and exactly the kind of thing more than three decades in rooms like that one teach you to feel rather than calculate.

I will admit something, because the honest version is more useful than the polished one. For a while I was as seduced by the speed as anyone, until it dawned on me, later than it should have, that the clock was rarely what lost the case.

There is one piece of evidence supporting this. Deloitte's 2026 corporate affairs report found that ninety five per cent of corporate affairs leaders say they are prioritising AI upskilling, and yet only about a quarter of them have a formal strategy for it. Read it the way I do and it is faintly comic and quietly alarming. Three quarters of the function whose entire job is managing what the organisation says have no agreed view on the technology that is now the first reader of everything the organisation says. We have noticed the weather and not yet bought a coat.

The question to start asking

So I would retire the question the industry has spent the year asking. Stop asking whether you can trust AI in a crisis. You can supervise your own use of it, you should, and that is a solved-enough problem. Start asking the question almost nobody is asking. Who is reading me through it, and have I written a single word for that reader?

Because the table has changed shape, and the old playbook, the one that assumed a patient human at the other end, is quietly drafting for someone who is no longer first in the queue. Get that right and the speed becomes what it should always have been, a convenience and not a strategy. Get it wrong, and you will draft the most reassuring statement of your career, and hand the claimant their best paragraph in it.

If you are weighing whether to settle or fight once a regulator has turned a dispute into a public crisis, that is the subject of Settle or fight: the decision that sets a regulator-driven crisis.

See the Reputational Resilience scorecard Speak with Tim in confidence

Tim Sutton is a senior reputational adviser to boards in their hardest moments and counsel to PR-agency principals on both sides of a transaction. timsuttonpr.com · LinkedIn