ANZ's CEO will not pretend he knows what AI does to the headcount

You've probably sat through a summit panel where a chief executive is asked what artificial intelligence will do to the workforce, and answers with a waterfall of roles evolved, roles automated and net new jobs, as if the rows had been audited. They have not. They are a wish. Nuno Matos, chief executive of ANZ, did the least fashionable thing a bank CEO can do at the Australian Financial Review Asia Summit. He said he does not know.

His line, quoted via Bloomberg, was that "I don't know and I think whoever says something that is certain will lie to you - you don't know. Nobody knows." Most CAIO packs treat uncertainty as a branding problem. Matos treated it as a fact.

What is on the record at ANZ

ANZ has around 40,000 staff. Last year it said it would cut about 3,500 by September 2026 to simplify the bank. That is a simplify programme. It is not, on the public record I am using here, an "AI did it" programme. Keep those two facts in different columns or you will write fiction.

Matos did not rule out AI-driven cuts. He also did not announce them. He pointed at the people building the models. AI, he said, is creating risks at a much higher pace than their own builders were thinking. He named Elon Musk, Dario Amodei and Sam Altman asking to slow down, and said that should tell us something clear: the risk is above what they expected. He also said AI can attack infrastructure without enough guardrails.

That is a chief executive talking about control, not a training catalogue. If you run learning and development, do not turn this into "ANZ needs more e-learn." He is talking about whether the bank understands the tool it is buying.

The rest of the street, with the caveats on

HCA Mag reports that Commonwealth Bank of Australia cut hundreds of contractor roles after putting AI on the customer chat. That is the magazine's account, not a CBA filing. The same piece says Standard Chartered will cut about 7,800 back-office roles by 2030, more than 15% of support functions. Those are their announcements, not ANZ numbers.

The pattern is familiar. Contractors go first. Then "simplify." Then someone writes a thought-leadership note about reskilling the people who remain. If you are the chief human resources officer, the honest version of this week is that we do not know the jobs number, and we do know the risk number is moving faster than the training number.

I have sat in rooms where the training is a 20-minute module and the risk paper is 80 pages. That ratio is the story. AI skills work that starts on a hated workflow beats a module titled "Introduction to Generative AI" that nobody fails.

What this means for L&D

If I were on Matos's communications team I would not "clarify" the quote. I would put the simplify 3,500 in one appendix and the AI-unknown in another. The market can handle two facts. What it cannot handle is a slide that says "our people will be upskilled" next to a contractor reduction you will not put your name on. ANZ has not, in this source, announced AI-driven cuts. Report that absence. It is news.

The quotes still need the same hedges: nobody knows, risk above what builders expected, infrastructure without guardrails. None of that is a training KPI. Do not let L&D turn a risk speech into an attendance report.

Put Matos's nobody-knows line on the first slide of the next steering pack. Then show what you are actually measuring this quarter: hours returned, error rates, contractor spend. If you cannot show those, you are not transforming. You are hoping.

Sources: HCA Mag / HRD, 16 September 2026, quoting Bloomberg from the AFR Asia Summit.