This spring marked EY’s Audit arm’s moment to embrace agentic AI, rolling the technology out across the 130,000-strong global operation active in more than 150 countries. 

Following the pilot stage, multiple AI agents are now embedded in the ‘EY Canvas’ assurance platform, which supports the delivery of more than 160,000 EY audit engagements and, according to the company, processes up to 1.4tn lines of journal entry data per year. This agentic integration is expected to support all end-to-end audit activities by 2028.

The first agents in audit, tasked with research, were initially released in 2025 before being rolled out across the global business in spring 2026 according to EY UK partner Richard Harrison, who oversees large-scale, complex audits. “So that was: ‘go help me understand technical accounting, go understand audit methodology’," he explains. 

“Then in spring this year, we moved towards project management administrative agents, and throughout the summer, we started releasing agents that help draft audit work, and [deployed] agents that support auditors with their reviews,” he says. 

Collectively, the agents improve the speed and accuracy of discovering anomalies, tracking trends and identifying risks within the data under audit, while accessing continuously updated auditing and accounting guidance.

The April launch was the latest in a series of plans announced four years ago to invest more than $1bn to transform its assurance services, when it began standardising its assurance enterprise technology systems across the world under one interconnected platform, residing on EY's private cloud.

‘Principles of responsible AI'

“Everyone's working from the same methodology. So, if you develop, test, validate, or certify a specific agent in one market, you can roll it out and scale it worldwide,” explains Harrison. 

With stories of AI agents ‘going rogue’, the only probable silver lining is that they now force firms to think deeply and early on about agentic use cases, safeguarding and deployment strategy. 

“Our agents are focused on very specific tasks, and operate within very clearly defined boundaries,” Harrison says. “So they only have access to specific information and tools relevant to the task they've been designated. They go through a very careful governance and testing process we call ‘certification’. We also have a responsible AI framework that sits around that.” 

At its core are what EY calls its “nine principles of responsible AI” that guide AI design, development and deployment. “And then you build an ‘orchestrator’ that sits over the top to get the right agent to do the right thing at the right time,” Harrison explains. 

Training, AI literacy and fluency

So how does EY get AI agents and human auditors to work in harmony? “Before we started rolling out enterprise-level technology, we started with training and productivity-based AI tools,” Harrison recalls. 

“So, teaching people about the art of the prompt: what a good or bad prompt is, what good AI output looks like versus non-transparent AI output. Getting people used to the concepts of the technology and how to interact with it. So how to use it, but then also the safeguards and responsible principles around that,” says Harrison, emphasising strict EY policy dictating which AI is applied to workflows.  

In training auditors, Harrison distinguishes between AI “fluency and literacy”. “AI literacy is: do you understand how it works? What it can do? How to use it? Whereas AI fluency is: do you understand how to talk about it? When you're having a conversation with a client around their own safe use of AI, how to assure AI, how we're using it and for what purpose, is that concept of fluency. I think that's valuable, for our trainees, and across every level.”

He adds: “Building AI fluency will be important for whatever our people need to do in their future careers, because AI tech is going to be, in my view, foundational to so many different things.”

We take that cue to ask how EY anticipates AI altering early-career auditors’ jobs. Will there be the need to search for a different type of graduate? 

Don’t lose sight of the fundamentals

Harrison isn’t convinced that AI agents will extinguish junior roles (in the way that other industries have lamented) and instead warns about “not losing sight of the fundamentals,” in the headlong rush to adopt AI and transform industries. 

“I think it'd be very easy to get over-focused on the new and forget that if you're a trainee auditor, you still need to learn accounting, still need to learn debits and credits, accounting methodology, how companies work, risk assessment. Those foundations are as important, if not more important than ever, because if you're going to be the human in the loop, you need to have that understanding.”

“Think about what makes a good finance auditor or finance professional. It's someone who's inquisitive, who's sceptical, independent, articulate. Someone who's motivated. I don't think any of that changes. What it [AI] is likely to do is accelerate people's career trajectory and really optimise certainly early stages of your career.” 

“But what you then need to do on top is train people to use the technology, who have the curiosity and mindset to challenge the output, while also training them how to deal with AI risk management, particularly when [client] companies reporting are using AI.”

“While aptitude and the ability to confidently use this technology is, of course, going to be important, the design also has to be easy to access, intuitive, and frankly, better to use than not to use,” says Harrison.

Ensuring there are humans in the loop

Most corporate AI strategies reference the nomination of ‘humans in the loop’ in a way that suggests a centralised approach to overseeing the work, scope of operations and consequences of deploying the technology on tasks. But for EY, each of the 130,000 assurance professionals across the world using agents on the EY Canvas platform is the human in the loop. Each is responsible for reviewing the outputs of any AI brought in to assist. That is, alongside the development teams who monitor agent performance and behaviour, including signs of model drift or other issues that could cause an agent to operate outside its intended scope.

EY has built human-centric, chat and messaging interfaces through which auditors can query agents’ processes; agent-generated audit trails that track sequences and explain the logic of their work, as well as runtime monitoring that measures model agent behaviour against the required benchmarks.

“So when the auditors are interacting with it, they can see exactly what it's done, how it's done it, what information was used, what's it matched, and then can validate it as part of that responsible loop,” says Harrison. 

With the critical nature of the audit and increasing regulatory pressure that demands human culpability for any enterprise AI failings, the prospect of agents ‘running the show’ at EY seems improbable.

“I can't see our people not ultimately being responsible for this in the foreseeable future,” Harrison agrees.

“You can't delegate authority to an artificial intelligence, and I don't think you'd want to. This is about how we use AI to support and enhance what we do, so that our people can focus on where they can add real value, perspective and sharpen our insights.”