LO
Auditor
Is the outside party who says someone else's numbers can be relied on — and whose signature is the only thing that makes that worth anything.
This is not a probability of losing your job. It combines how much of the role's task load is exposed to automation with how far adoption has actually gone — useful for comparing occupations on one consistent basis, and for nothing else.
Written for external audit and assurance in a firm, and largely applies to internal audit too. It is a different occupation from accountant: an accountant produces the numbers and closes the period, an auditor is the outside party who tests them and signs. The tax practice is different again. Regulation matters more here than on most pages — who may sign, and what they are liable for, is set by law and differs by market.
The evidence base holds verified records for other occupations, but not one for this one yet. Until it does, the analysis below is reasoning about task structure and known technical capability — for this job in particular it is not backed by traceable sources, and we would rather say so than cite things we have not verified. An empty section here is a gap in our coverage, not a finding about the work.
What is actually changing#
The unit of analysis is the task, not the job title. A role is not replaced — its task mix shifts.
Is this your job? Say so and this page narrows to your share of it.
A job title is a bundle of tasks bought together, and no two people hold the same bundle. Nothing is sent anywhere — it stays in this browser.
Sampling and testing
Automating≈ Platform inferencePulling a sample of transactions and checking them against the supporting documents.
Sampling exists because testing everything was too expensive. Once testing everything becomes cheap, the sample loses its reason to exist — and full-population testing has been ordinary in large audits for a decade, well before anything called AI. What generative tools add is the ability to read the supporting document rather than only the ledger entry, which extends the same trend into the half that used to require eyes.
Says nothing about how audit hours are actually billed, and in many firms the hours freed here were absorbed into more testing rather than into fewer people. It also says nothing about smaller audits, where the tooling is often not bought at all.
Deciding where to look
Still human-led≈ Platform inferenceRisk assessment: which accounts, which estimates, which part of this business could hide something.
Anomaly detection genuinely helps and finds things a person would miss. What it cannot do is know which unusual thing is worth pursuing in this business, this year, given this management's incentives — that requires a model of why someone would want to misstate, and the incentive lives outside the data.
A judgement about the nature of the work, not a measurement. It also does not claim auditors are good at this: missing the risk that mattered is the defining failure of this profession and there is a long public record of it.
Asking the question that gets a real answer
Still human-led≈ Platform inferenceSitting across from a finance director and working out what is not being said.
Evidence in an audit is not only documentary; a great deal of it is what someone does or does not say when asked directly. That is a live social act with a person who may have a reason to mislead, and the auditor's professional scepticism is exercised in the room, not in the file.
Nothing here measures how much of a modern audit is inquiry versus document testing, and in many engagements the inquiry is a formality completed by email. Where it is a formality, this task is not human-led — it is simply not being done.
The file and the opinion
Being augmented≈ Platform inferenceDocumenting what was done, and drafting the wording that goes in front of shareholders.
Working papers are structured, repetitive and heavily templated — close to the ideal case for generation, and this is where firms claim most of their time saving. The opinion itself is different in kind: its wording is regulated, and a modified opinion is a public act with consequences for the client that the person signing has to be willing to cause.
No verified record on this site measures documentation time saved in audit, and firm claims about it come from a party selling the transformation. What would settle it: a regulator's inspection findings on files prepared with these tools.
Signing it
Still human-led≈ Platform inferenceBeing the named person whose licence stands behind the opinion, and who answers if it was wrong.
Not a difficulty claim. In every market that has a statutory audit, the law requires a licensed individual or firm to sign, and attaches liability to that signature. That is the whole product: an audit is valuable precisely because someone can be sued over it. Automation does not reach a position that exists in order to be liable.
It protects the signature, not the headcount behind it. A firm can sign the same number of opinions with fewer people, and the signature says nothing about how many juniors were needed to get there — which is the number most readers of this page actually care about.
Owning what the machine drafted
New task≈ Platform inferenceChecking machine-drafted professional output before it leaves under the firm's name — citations, quotations, the things that look right.
New work created by the tooling, and the failure mode is specific: generated professional text is most dangerous where it is most confident, because a fabricated citation looks exactly like a real one to a reviewer who is skimming. This site holds a record of a Big Four firm repaying part of a government fee after fabricated references and a misattributed court quotation were found in a report it had drafted with a generative tool.
That record is one consulting engagement at one firm, not a statutory audit, and it establishes that the failure happened rather than how often it does. New work appearing is also not new headcount: in a billable-hours business an unbilled checking duty is absorbed, which is exactly the condition under which it gets skipped.
Which technologies matter here#
Four separate signals. They are deliberately not added together — a job exposed to two technologies is not twice as exposed.
How it got here#
The index is not a static number. This is where it would have sat at each capability checkpoint since ChatGPT — reconstructed, and labelled as such.
A high start that is not about generative AI: full-population testing was already ordinary in large audits a decade ago, so the profession entered this period with its most visible task — sampling — already losing its reason to exist. Sampling is unusual in that it does not become obsolete by being done better; it becomes obsolete by becoming unnecessary once testing everything is cheap. The 2023-2025 rise is generation reaching the working papers, the most templated documentation in professional services. It flattens because of something written into law rather than into the technology: in every market with a statutory audit, a licensed person must sign, and liability attaches to that signature. Note what the curve protects and what it does not — the signature, not the number of juniors needed to reach a signable file.
A flat line is not a forecast of safety. It says which tasks automation has reached so far — the occupations that moved least here are the ones where the constraint is physical or regulatory, and both of those can change.
Recent changes#
No verified events recorded yet.
This section will fill from the monitoring pipeline as events are collected, de-duplicated, graded and linked to the tasks above. An empty list here means we have not verified anything — it does not mean nothing is happening.
"We found no news" is not the same as "you are safe."
What this means for you#
The work you would be hired to do first — pulling samples and tying them to documents — is the task whose reason to exist disappears when testing everything becomes cheap. That is worth knowing before you spend three years getting fast at it. What holds is knowing where to look and being willing to ask the question nobody wants asked, and neither is on the exam.
The signature is protected by law, and that is a real and unusual kind of protection — but it protects the signature, not the number of people behind it. The honest question for a firm is how many juniors are needed to reach a signable file, and that number is the one the tooling is aimed at. If your value is that you sign, you are safe and your team may not be.
Your options#
Four directions, each with its real constraints and one thing you can test this week. Continuing as you are is a legitimate choice — it just has to be a chosen one.
Own the risk judgement, not the testing
When everything can be tested, the scarce act is deciding which unusual thing is worth pursuing — and that needs a model of why someone would misstate, which lives outside the data.
It is judgement you are only trusted with after years of the testing work that is being automated away, which is a real and unresolved pipeline problem for this profession.
Take last year's file on one client and find the risk you decided not to pursue. Write down why, and whether you would decide the same today.
Be the one who audits the model
Companies are putting models into processes that produce numbers, and somebody has to test a control that is a model rather than a procedure. That is this profession's own skill applied to a new object.
The standards for it are still being written, so you would be forming the method rather than applying one.
Find one control at a client that is now a model output and ask how its performance is monitored. If the answer is nobody's job, that is the work.
Cross to the side that gets audited
Knowing precisely how an audit finds things is worth a great deal inside a finance function, and the move is well-trodden and well-paid.
You stop being independent, which is the whole basis of what you were selling — it is a different kind of value, not the same value indoors.
Ask a client's finance director what they wish their auditor understood about their business. The gap in the answer is the job.
Common questions#
The parts move in opposite directions, so one number would hide the thing you need. Sampling has been losing its reason to exist since long before generative tools — full-population testing is a decade old — while the signature is protected by statute in every market with a statutory audit. A signal you can check yourself: how many people were on your last engagement compared with the same client three years ago, and whether the hours moved to testing more or to fewer juniors. The tooling is aimed at that second number.
Deciding which of the things it flags is worth pursuing, asking the question that gets a real answer, and being the person whose licence stands behind the conclusion. It is worth being precise about why the last one holds: it is not that signing is hard, it is that an audit is valuable precisely because somebody can be sued over it. A position that exists in order to carry liability is not reached by a better model.
The record on this site says the failure happened; it does not say anyone stopped. Read it for the mechanism instead, because the mechanism is the useful part: generated professional text is most dangerous where it is most confident, since a fabricated citation looks exactly like a real one to someone skimming. The control that catches it is a person reading carefully — an unbillable duty in a billable-hours business, which is precisely the condition under which it gets skipped.
No, and the difference is the point. An accountant produces the numbers and closes the period; an auditor is the outside party who tests them and signs. The most exposed task is different on each: on the accountant page it is entry and reconciliation, here it is sampling — and sampling is unusual because it does not become obsolete by being done better, it becomes obsolete by becoming unnecessary. If you hold both roles, which many people in smaller markets do, read both pages and expect them to disagree about what protects you.
Studying towards this?
These majors lead here. Their pages break down which of their competencies transfer and what graduates typically lack.
Method and sources#
- Assessment date
- 2026-09-12
- Basis of the task judgements
- 0 evidence-backed · 6 platform inference · 0 not enough evidence
- Verified events
- 0