NextFin News - Nuno Matos, chief executive of Australia's ANZ Group Holdings Ltd., has warned that artificial intelligence is generating risks "at a much higher pace than what their own builders and developers were thinking," and declined to rule out large-scale job cuts at the bank as it adopts the technology. The comments, delivered at the Australian Financial Review Asia Summit in Sydney on Tuesday, place one of the country's Big Four banks squarely inside a widening global debate over how fast frontier AI should advance - a debate that over the weekend drew OpenAI's Sam Altman, Elon Musk, and Anthropic's Dario Amodei into an unusual public chorus calling for the industry to slow down.
The Warning From Sydney
Matos spoke at the AFR Asia Summit, held at The Fullerton Hotel Sydney on September 15, 2026. His message was blunt: the speed at which AI capabilities are advancing is outpacing the ability of the people building these systems to anticipate the risks they create. "At this point in time, it's creating risks at a much higher pace than what their own builders and developers were thinking," Matos said.
The ANZ chief did not stop at risk. When pressed on the workforce implications of the bank's AI adoption, he declined to rule out large-scale job cuts - a stance that carries weight because Matos has already set in motion one of the largest restructurings in Australian banking history. In September 2025, ANZ announced it would eliminate approximately 3,500 staff positions and around 1,000 contractor roles by September 2026, a reduction of roughly 8% of its workforce, at a one-off restructuring charge of A$560 million (about US$370 million).
The timing matters. Matos's AI warning comes just days after three of the most powerful figures in the technology industry - Altman, Musk, and Amodei - publicly backed calls to slow the pace of AI development. Amodei, Anthropic's chief executive, wrote in an essay over the weekend that AI companies "must slow the pace at which we improve the capabilities of AI models." Altman followed on X, welcoming "a federal framework that sets consistent safety requirements for frontier AI" and warning that "no amount of American competitive pressure should justify recklessness." Musk, long a critic of uncontrolled AI development, aligned himself with the slowdown position. The technology sector's own architects are now publicly worried about the speed of their creation; a banker saying the same thing carries a different kind of weight.
For a bank chief executive to echo that caution is notable. Financial institutions are among the heaviest investors in AI, deploying it across customer service, fraud detection, credit assessment, and back-office operations. When the CEO of a systemically important lender says the technology is moving faster than its creators can manage, it signals that AI risk is migrating from the technology sector into the core of the financial system. Banks are not just users of AI. They are its exposure point to the real economy.
Why a Bank CEO Is Talking About AI Risk
The link between AI and banking is not abstract. Banks run on information processing - assessing creditworthiness, detecting fraud, pricing risk, serving customers, and complying with regulation. Every one of those functions is a target for AI automation. Goldman Sachs Research has estimated that roughly 300 million full-time jobs globally are exposed to AI automation, with about 25% of all work hours in the United States potentially automatable using technology that exists today. Administrative and legal professions show the highest exposure, but financial services roles sit squarely in the crosshairs. A bank's core product is the assessment of risk, and AI is a technology for assessing risk faster and cheaper than trained analysts.
McKinsey's research points in the same direction: half of today's work activities could be automated in scenarios ranging from 2030 to 2060, with a midpoint of 2045 - roughly a decade earlier than the firm's 2017 estimate. For banking, the implication is not a distant theoretical shift. It is a planning horizon that aligns with the multi-year transformation programs already under way at ANZ and its peers. The technology is arriving inside the window of a single CEO's tenure, which is why Matos is treating it as a current management problem rather than a future strategic consideration.
Matos has been candid about how ANZ intends to use the technology. In an investor briefing in May 2026, he said the bank had launched agentic AI-enabled capability in its customer relationship management system and was up-skilling business bankers through an upgraded Banker Academy, with a commitment to increase the number of business bankers by close to 50% by 2030. The bank also said it would maintain investment spending at about A$1.5 billion a year. That framing - AI as a tool that augments frontline staff rather than simply removes them - is the public face of the strategy. The job cuts are the cost of getting there.
That distinction matters, and it separates the cyclical leg of this story from the structural one. The cyclical leg is the cost-cutting cycle: ANZ's statutory profit for the year ended September 2025 fell 10% to A$5.891 billion, the bank is the smallest of Australia's Big Four by market valuation and its share price has lagged rivals, and Matos is clearing the decks. Investors have been willing to back him, but the market has treated the restructuring as a necessary reset for a laggard, not as evidence of a permanent shift in how banks employ people.
The structural leg is different. Once AI capabilities reach a threshold where they can reliably perform cognitive work that previously required trained humans, the demand for certain categories of white-collar labor does not bounce back when the cycle turns. A bank that can automate document review, compliance monitoring, or routine credit analysis does not rehire those analysts when profits recover. The work has been permanently transferred to software. That is why Matos's refusal to rule out further job cuts is more significant than the 3,500 roles already announced: it treats displacement as an open-ended process, not a one-time restructuring event. The 3,500 cuts are the first tranche; the AI-driven reductions are the open-ended second act.
The competitive pressure is visible across the sector. Commonwealth Bank of Australia has run a A$90 million program to prepare its workforce for an AI-driven workplace, and its chief executive, Matt Comyn, wrote in May 2026 that AI "will have workforce consequences throughout the economy" and that "some roles will be reduced in number." Westpac has planned cuts of around 1,500 roles. National Australia Bank's leadership has said job cuts are inevitable as AI reshapes the workforce. When every major bank is moving in the same direction at the same time, the cuts are not a competitive choice. They are an industry-wide repricing of the value of white-collar labor.
The Second-Order Question the Market Isn't Asking
The first-order reading of Matos's comments is straightforward: AI will cut bank jobs, and AI carries risks that executives are starting to fear. That reading is already priced into the sector. Australian banks have been cutting roles for two years, and the market has treated those cuts as a positive for profitability. ANZ's shares have responded to the restructuring as a necessary step toward closing the performance gap with its rivals.
The second-order question is harder: what happens to the risk profile of the financial system when AI moves from a support tool into core decision-making? A bank that automates credit assessment, fraud detection, and compliance monitoring is not just reducing headcount. It is concentrating decision-making in models whose failure modes are, by Matos's own description, moving faster than the builders' understanding of them. If those models fail in correlated ways - during a stress event, across multiple institutions using similar technology - the result is not a staffing problem. It is a stability problem.
This is the transmission channel that separates a cyclical cost story from a structural risk story. Job cuts are a one-time earnings event, reflected in a restructuring charge and a lower cost-to-income ratio. Model risk embedded across the system is a persistent, compounding exposure that does not show up in quarterly earnings until it fails. The market is good at pricing the first. It has no established framework for pricing the second.
Regulators know this. Australia's transaction-crime regulator, AUSTRAC, warned in September 2025 that the wave of banking job cuts - almost 6,000 roles across the sector at the time - must not weaken risk management and compliance, which were strengthened after the banking royal commission. AUSTRAC chief executive Brendan Thomas threatened to escalate regulatory oversight if the job losses led to weaker compliance or reduced investment in anti-financial-crime controls. That warning sits awkwardly alongside the industry's AI enthusiasm. The same functions regulators want protected - compliance, financial-crime monitoring, risk management - are precisely the functions banks are most eager to automate. If AI makes those functions cheaper and faster but less interpretable, the regulator's concern and the bank's efficiency drive are on a collision course. Matos's comment that risks are building faster than builders anticipate is, in effect, an admission that interpretability is the constraint.
The Counter-Thesis: AI Creates More Jobs Than It Destroys
The strongest argument against the job-cut narrative is that AI raises productivity, lowers costs, and creates new categories of work faster than it destroys old ones. Optimists point out that previous technological waves - from spreadsheets to the internet - eliminated specific roles while expanding employment overall. Some technology leaders have walked back their own earlier doom predictions. Sam Altman himself said in May 2026, speaking at a Commonwealth Bank of Australia conference in Sydney, that he no longer expected a "jobs apocalypse" and that AI had not claimed as many white-collar jobs as he had feared. "I now think I understand more about why it hasn't," Altman said, "and I'm obviously grateful but that is an area where my intuitions were just off."
That counter-thesis is credible on a long enough horizon. But it does not answer the timing problem that Matos and other bank CEOs face. Even if AI is net-positive for employment over a decade, the displacement happens at the speed of software deployment, while retraining and redeployment happen at the speed of human institutions. A bank that can automate a function this quarter does not wait ten years for the new jobs to materialize. The cuts are front-loaded; the creation is back-loaded. For the 3,500 ANZ employees exiting by September 2026, and the thousands more across Australia's banking sector, the long-run equilibrium offers little comfort.
There is also a distributional problem the optimists' aggregate numbers hide. The jobs AI displaces in banking - operations, processing, routine analysis, parts of customer service - are not the same jobs AI creates. The new roles demand skills in data, model oversight, and engineering that a displaced operations analyst does not automatically possess. Goldman Sachs has noted that workers displaced from knowledge industries may be less suited to the kinds of labor that are most needed, a mismatch that shows up as structural unemployment even when aggregate employment holds. The economy can be at full employment while a specific cohort of workers is permanently displaced.
The falsifying signal for the structural-displacement view is specific and observable: if, over the next two years, Australian banks that aggressively adopt AI show rising total headcount alongside flat or falling labor costs per unit of output - meaning productivity gains are being reinvested in hiring rather than extracted as cost savings - then the job-cut thesis is wrong. If instead headcount falls while labor costs per unit of output also fall, the displacement is real and structural, not cyclical. The first data point to watch is ANZ's own headcount disclosure in its next full-year results, due after its September 2026 year-end.
What Comes Next
Three signals will determine whether Matos's warning becomes a sector-wide playbook or remains an isolated caution. First, the detail in ANZ's own AI deployment: whether the bank frames AI as a headcount-reduction tool or a capability-reinvestment tool, and whether its headcount trajectory matches its rhetoric. Second, the regulatory response: whether AUSTRAC and the prudential regulator impose constraints on how quickly banks can automate risk and compliance functions. Third, the technology itself: whether the pace of AI capability growth actually slows, as Amodei and Altman are now urging, or accelerates past the point where voluntary restraint holds.
For investors, the read differs by time horizon. In the short term, cost cuts support margins, and the market has rewarded banks that execute them; ANZ's restructuring is expected to deliver savings that flow through to profitability as roles exit. In the medium term, the risk shifts to execution and regulation: banks that automate risk functions fastest may also be the first to discover what Matos is warning about, and regulators have already signaled they are watching. In the long term, the direction is structural: the cognitive work that banks have employed people to do for decades is now technically automatable, and that does not revert.
The scenarios are clear. In the base case, banks automate routine cognitive work gradually, headcount declines modestly, margins improve, and regulators impose light-touch oversight that slows but does not stop deployment. In the upside case for labor, AI proves harder to deploy safely in regulated environments than vendors claim, interpretability requirements force humans to remain in the loop, and headcount stabilizes as new AI-adjacent roles absorb displaced workers. In the downside case, model failures at one or more major institutions trigger a regulatory crackdown that forces costly remediation, and the cost savings investors have priced in fail to materialize. The variable that determines which scenario plays out is not the technology. It is whether the people overseeing it can understand it well enough to control it - the exact question Matos says they currently cannot answer.
"At this point in time, it's creating risks at a much higher pace than what their own builders and developers were thinking."
The market has priced the cost savings. It has not priced the risk that Matos is describing - a financial system where the models making decisions are evolving faster than the people overseeing them can understand. That is the gap between what the AI optimists are selling and what a bank chief, staring at his own balance sheet, is starting to see.
Explore more exclusive insights at nextfin.ai.
