Accountancy is not the most glamorous of professions, but some excitement has crept into the job lately. Technology is changing the role of auditors to make the work much more interesting than when I started, Catherine Burnet, KPMG's UK head of audit, told me last week. The thrill is being caused by the rapid adoption of AI at the Big Four accounting firms, both globally and in the UK.

The work of auditing large companies is being transformed by AI agents filtering transactions in search of fraud and errors, reviewing the work of junior auditors, and generally making themselves useful. Matthew Campbell, KPMG's chief technology officer for UK audit, compared it to filtering an entire river rather than dipping a bucket into it to gather a sample. Audit firms that face the danger of failing to identify problems in a mass of data naturally find it attractive to be able to deploy an army of AI agents alongside humans.

EY employs 85,000 auditors globally but conducts nearly twice as many company audits each year. AI could prevent auditors becoming overwhelmed. They could do with some help.

Firms in the UK have been under pressure to improve audit standards since a series of failures including the 2018 collapse of the building services company Carillion. KPMG was fined 21 million in 2023 for its Carillion failures. They have made progress, helped by technology: both KPMG and EY now use their own cloud platforms to ensure audits are more thorough.

Corporate clients are using AI to organise and track their finances, and accountants do not want to be left behind. But audit firms should not get carried away. For one thing, AI changes nothing in terms of regulatory accountability.

A firm not only has to stand behind its audit opinions, but be able to explain the reasoning and justify the action it took on each transaction it scrutinised, or asset value it checked. It cannot point the regulator to a black box. In one sense, that is the point of AI filtering all transactions and identifying higher-risk cases to be checked further: the most sensitive work remains with humans.

But the AI model is triaging, not auditing the material in depth, and how can the firm be confident that its reasoning is sound? It could be producing a string of false negatives by failing to spot problems. This speaks to a wider difficulty: the human inclination to place too much faith in technology. Banks' over-reliance on value-at-risk models that underplayed the chance of severe losses in a real estate crisis was a factor behind the global financial crisis in 2008-09.

AI audit models could work smoothly for a long time, then one day be fooled by a novel kind of transaction. Richard Harrison, EY's UK assurance digital leader, says its principle is that, AI is here to support us, not to replace us. It is trying to ensure that its AI agents prompt auditors, not dictate answers to them.

Both audit firms and the Financial Reporting Council, the UK regulator, say they want AI to help them improve audit quality rather than focusing on cost-cutting through the elimination of junior jobs.