They were betting
nobody would check.

The job is a bet

Start with what these products actually are.

A statutory audit tests samples, and says so plainly — reasonable assurance, never absolute. A penetration test covers a scope, not your estate. A certification reviews selected controls on a single date. Nobody has ever checked everything, and no honest practitioner claims otherwise.

That is not fraud. It is arithmetic. Checking everything was never economically possible, so the profession sold the best thing it could produce, and wrote the limits into the wording.

But it means every signature has carried a second clause that nobody says out loud. Not only I checked what I could — but nobody will ever check the rest.

Why the bet was safe, and why it paid

The gap between the sample and the whole was protected by one thing: cost. Re-performing the work completely would have cost more than the fee, more than the lawsuit, more than anyone was willing to spend. So the gap was never tested, and an untested gap is indistinguishable from no gap at all.

Now add the other half, which is the part people miss. Demand for this paper is compulsory. The regulator requires the audit. The customer's procurement form requires the certificate. The insurer requires the report. You cannot decline to buy it.

Guaranteed demand, and an output whose quality literally nobody could measure. That is one of the best businesses ever constructed, and it has made a great deal of money. Not by cheating. By selling into a market with no possible quality check.

The first card turned over

Last month an operator pointed a swarm of AI agents at two flaws in print-management software and broke into at least 395 organisations across 48 countries. One school went from first contact to full administrator in seven minutes.

The flaw was public. The vendor had shipped an emergency patch three days earlier. The software ran with the highest privileges on the machine by default, because that is how it installs. And where somebody had put an ordinary web firewall in front of it, the attack was simply blocked.

Plenty of those 395 will have had a clean report in a drawer. Nobody had to prove the paper was worthless. The breach did it for them — and it did it at a cost of nothing, in seven minutes, without a single lawyer.

Checking just got cheap

That is one card. Here is why the whole deck is now exposed.

The same operator went from an empty workspace to working exploitation in four hours. Read that as a price, not a threat. If a stranger can build a complete, repeatable check of your defences in an afternoon, then a complete check is no longer expensive — and the cost that protected every gap in every report has collapsed.

I made this argument about tax recently: enforcement was priced on the cost of looking, and when looking got cheap, sampling became a census. Assurance is the same trade running the other way. When a full check costs minutes instead of months, “we sampled and found nothing” stops being a defence.

And something else changes with it. You cannot prove negligence against a method nobody is able to re-run. You can prove it against one anybody can.

The second card

Which brings me to the thing that has bothered me all week. I am going to speculate here, so read it as speculation.

A local audit firm resigned as external auditor of a string of listed companies inside the same fortnight. I am not naming the firm or the companies. The filings state the resignations were voluntary and made on the ground of broader portfolio and resource-management considerations, and nothing in them suggests a problem with anybody's accounts. I know nothing beyond what was filed, and the boring explanations are the likeliest — fees, partners leaving, a book outgrowing its staff. Auditors resign more often than people assume.

But it made me ask a question I could not put down. What does a signer actually do, when the box starts becoming openable?

He stops signing. Resignation is the cheapest risk management ever invented — you cannot be sued for an opinion you never gave. Walking away costs a fee. Being wrong in writing costs everything. And it is close to the only method this industry has ever needed, because until now nothing forced a firm to price its own exposure.

And a second blade is closing at the same time, which is what turns walking away from cautious into clever. The legal shield is being narrowed. One of the major audit regulators has just moved the standard for individual practitioners from recklessness down to plain negligence — meaning a person can be pursued for work done in good faith but carelessly. That is precisely the work sampling produces. Practitioners objected during the consultation, and you can see why.

And this has stopped being theoretical. In September 2024 China’s finance ministry and securities regulator suspended PwC Zhong Tian, the firm’s mainland arm, for six months and fined it 441 million yuan — about US$62 million, the heaviest penalty ever imposed on a Big Four firm in China. The finding concerned its audit of Hengda Real Estate, Evergrande’s onshore flagship, and the bond issues of 2019 and 2020. The regulator concluded the work was negligent, and that 88 per cent of the firm’s observation records on Evergrande’s property projects across those two years were “inauthentic or untrue.”

Read that percentage again. Somebody finally opened the box, and most of the work was not inside it. Hong Kong’s regulator fined the firm separately, and more than fifty clients walked within months — the market punished it harder and faster than the fine did.

Then, this month in Singapore, liquidators of four companies filed a S$1.30 billion claim — roughly RM4.2 billion — against DBS, Singapore’s largest bank, over accounts those companies held there in 2013 and 2014. Malaysia’s asset-recovery taskforce says more than US$1 billion of money misappropriated from 1MDB and SRC International moved through them. DBS rejects the claim, says it will defend the case vigorously, and has made no provision for it. Nothing has been proven against the bank, and a filed claim is an allegation, not a finding. A separate US$2.7 billion suit against Standard Chartered over its own 1MDB role was filed last year.

Notice what is being argued in both. Not that the gatekeeper stole anything — that the gatekeeper who processed it should answer for not asking. The principle is reaching past auditors to anybody who waved something through.

As far as I can find, Malaysia has not passed an equivalent for company auditors, and I am not going to pretend otherwise. But the direction is not ambiguous, and it is the same direction everywhere. The world has had enough of “this is not a guarantee.”

Put the two together and every signer's position changes at once. Cheap verification makes bad work provable. A negligence standard makes it actionable. For decades the safe move was to sign and rely on nobody checking. The safe move now is to sign only what you would defend.

Nowhere to hide but out

So here is where I think this goes, and it is a prediction rather than a fact.

The firms that were quietly relying on nobody looking have two moves. They can leave — quietly, on the ground of portfolio and resource-management considerations. Or they can do the thing that was never affordable before: stand behind the work.

Because the same collapse in cost that exposes them also pays for their replacement. When continuous verification costs a fraction of a human team, a firm can finally afford to carry the risk it certifies — not an annual report, but continuous assurance with a number attached if it is wrong. The savings from automating the testing fund the warranty. AI does not only make attacks cheaper. It makes underwriting possible.

That is the honest fix, and every other serious profession got there long ago. An engineer answers for the building; a doctor carries indemnity with a number on it. Neither guarantees an outcome — the surgeon does not promise you live. They are liable for working negligently, and that single obligation is what makes a signature worth anything.

And it solves the problem a buyer cannot solve himself. A five-thousand-ringgit "penetration test" is a scan with a logo on the cover; a real one is people-weeks; the brochures look identical. Nobody outside the trade can tell them apart — but anybody can read a liability figure. The good practitioners, who are the real victims of this market, would be the first to benefit.

The bottom line

Four questions before you sign the next assurance invoice.

What is outside the scope, and who decided?
Is a retest included?
What happens between tests?
And what do you carry if we are breached anyway?

Ask the fourth and watch the room. Today most will carry nothing, and they will tell you so in careful language.

They were never selling certainty. They were selling a bet that the cards stay face down. The cards are being turned over now, and the ones who cannot stand behind their own signature have exactly two options: get real, or get out quietly while it still looks voluntary.