Insights

Westpac reported an annualised 150,000 banker-hours saved in the same window CBA's CFO conceded AI investment still exceeds the benefits realised, and CBA said it chose to reinvest the freed capacity into delivering more technology change rather than reducing resources. That is the mechanism reconciling two numbers that look contradictory: the hours are real and they are being recycled into throughput, not harvested as headcount. It is also why the job-cut question keeps being deferred rather than answered.

Example: CBA's record FY26 profit and its first hard AI payback numbers

The licensing regimes did not contain stablecoin issuance — they made it safe for incumbents to enter. Once reserve, redemption and issuer standards were settled across seven economies, bank issuance followed within months: U.S. Bank went live on a public chain, 21 institutions committed to a joint issuer, and 39 US state banking associations formed their own alliance. MAS's rule banning interest to holders is the tell, drawing the line that lets a bank issue a stablecoin without cannibalising its own deposit base.

Example: 21 major banks commit to a joint USD stablecoin venture

Netsafe and GASA put New Zealand's all-channel scam losses at $2.3bn against $265m reported by 12 banks — most of the harm never touches a bank's systems. Yet the only enforceable regime in the region runs through financial-sector machinery: Australia's Scams Prevention Framework made AFCA membership a gate on 1 September 2026, with substantive duties from 31 March 2027. Australia did extend those duties to telcos and platforms, which is the part New Zealand's voluntary, bank-only Code of Banking Practice has no answer to.

Example: NZ digital scam losses reach $2.3bn against $265m bank-reported

The dates that actually bind on post-quantum migration arrived through procurement and vendor contracts rather than prudential rulebooks — SWIFT's 2027 SwiftNet target and the 21 September FIPS 140-2 sunset, which cuts off legacy modules for new US federal purchasing. Supervisors then converged on the same seam from the other side, with the Bank of England telling firms to get assurance from material suppliers and the US Treasury's task force naming third-party readiness as a workstream. For a New Zealand bank the question lands as a vendor-contract review, not a security-architecture project.

Example: FIPS 140-2 sunset and BoE pressure on third-party PQC readiness

The same twelve months added a proposed NZ$209m prudential levy and a locked-in NZ$27.3m-a-year AML/CFT levy to New Zealand deposit takers' cost base — both recovering the cost of supervising the business banks already do. Over the same period the infrastructure their competitors are building acquired no local rulebook at all: the FMA's own tokenisation submissions report records submitters asking for the legal and prudential status of tokenised deposits to be clarified. Cost is rising against the existing perimeter while the next one stays undrawn.

Example: RBNZ opens consultation on a new prudential levy

Two acquisition signals landed within weeks. Toi Foundation's trustees chose Heartland's $620m offer for TSB over a late rival bid from Kiwibank's Crown-owned parent — meaning the Government was willing to see its own bank buy scale. Then ANZ was reported weighing a takeover of Judo Capital. The competitive answer to the challenger tier is turning out to be purchase rather than price, and both an Australian-owned major and a state-owned parent are now on the buy side.

Example: ANZ reportedly eyes takeover of business-lender Judo Capital

U.S. Bank put a bank-issued stablecoin into live production on a public blockchain in the same month the RBA restated that there is no policy case for a retail CBDC and redirected its work to wholesale tokenisation. The US Federal Reserve is statutorily barred from a retail CBDC until the end of 2030, and RBNZ's own digital cash sits around 2030. Private bank money is arriving on public infrastructure years before public digital money arrives at all — which settles, by default rather than decision, who owns the retail digital-payments relationship.

Example: RBA rules out near-term retail CBDC, doubles down on wholesale tokenisation

The AI-productivity claim and the AI-headcount claim have decoupled in banking. The dated, confirmed job cuts are landing on banks' suppliers and the card schemes, while banks convert AI time savings into throughput rather than headcount — which means the sector's job-loss reckoning is being deferred, not delivered.

• · Visa is cutting about 2,600 roles, roughly 7% of its workforce, concentrated in technology and product, with the CEO framing AI as reshaping how work gets done.
• · CBA disclosed around A$200m in gross AI benefits for FY26 while its CFO said investment had exceeded benefits across FY25 and FY26.
• · Scotiabank claims roughly 24,000 days of work saved in four and a half months and TD cut mortgage pre-processing from 15 hours to three minutes, but analysts note the gains have not appeared as headcount reductions.
• · ANZ committed $20m across Queensland and $10m across NSW to branch upgrades, relocations and expansions.

Counterfactual — If banks were substituting AI for labour at the rate their own productivity disclosures imply, CBA would not be reporting AI spend still exceeding benefits two years running while explicitly choosing to reinvest freed capacity into more technology change, and the institutions publishing the largest time savings would also be publishing falling headcount. Equally, if the claim were wrong, the confirmed large-scale cuts would be landing inside banks rather than at Visa, Mastercard and Block.

Stablecoin licensing regimes did not constrain the instrument — they legitimised it for incumbents. The arrival of settled reserve, redemption and issuer standards is what made bank-issued tokenised money possible, and banks moved within months of the rules landing.

• · Seven major economies now mandate full reserve backing, licensed issuance and guaranteed redemption, with the market around US$308bn.
• · U.S. Bank launched USBDC and completed a live cross-border pilot payment on Stellar while keeping the token inside its core finance, risk and compliance systems.
• · Twenty-one major institutions committed to forming a joint company to issue a USD stablecoin targeting the first half of 2027.
• · MAS's proposed framework bans paying interest to stablecoin holders specifically to keep stablecoins distinct from bank deposits.

Counterfactual — If licensing were constraining the instrument, bank issuance would be slowing as regimes bedded in and issuance would concentrate in the least-regulated jurisdictions. Instead the issuance is coming from the most heavily supervised institutions in the most regulated markets, after their rules took effect.

New Zealand's banks face a rising bill for supervising the business they already do, on a rulebook that does not yet cover the infrastructure their competitors are building. This is a domestic rulebook gap, not a capability gap in their Australian parents.

• · RBNZ opened consultation on a prudential levy recovering roughly NZ$209m over three years, with deposit takers carrying 54%.
• · Cabinet locked in an AML/CFT levy recovering about NZ$27.3m a year, with banks and deposit takers funding 85% of it from 1 July 2027.
• · No Australian or New Zealand institution is a founding member of the 21-firm USD stablecoin venture.
• · New Zealand still has no dedicated stablecoin licensing regime, and the NZDD token's 'not a financial product' status remains contested.

Counterfactual — If New Zealand were keeping pace, the legal and prudential status of tokenised deposits would be settled here rather than being the thing submitters are still asking regulators to clarify. And if the gap were really a capability gap in the parent banks, Australia's majors would have no tokenisation programmes — but they do, which localises the problem to the New Zealand rulebook rather than to the banks.

Most scam harm now originates outside banks' systems, but Australia has built the only regime in the region that puts telcos and platforms on the same dated obligations as banks — leaving New Zealand's voluntary, bank-only code increasingly isolated as the outlier rather than the norm.

• · Netsafe and GASA put New Zealand's all-channel scam losses at $2.3bn for the year to 30 August against $265m in bank-reported fraud across 12 banks.
• · UK Finance found scam attempts up 62% and phishing up 140% year on year across more than 100 million accounts.
• · AFCA has proposed nearly doubling its scam-loss compensation cap from A$631,500 to A$1.263m.
• · New Zealand's reimbursement obligations still sit in the voluntary NZBA Code of Banking Practice rather than in statute.

Counterfactual — If liability were genuinely being shared across sectors everywhere, the UK's platform-facing duties would already be in force rather than in consultation. And if the cross-sector model were purely rhetorical, Australia's framework would apply only to banks — which research shows is not the case, and that is the part of this claim that had to be narrowed.

Agentic AI's binding constraint in banking has shifted from model capability to data and systems architecture, which reclassifies it from an AI problem into a core-modernisation capex problem — and explains why the pilot-to-production gap persists across banks using identical frontier models.

• · Only 16% of banks have fully deployed agentic AI in card issuing against 52% piloting, with legacy systems built around human coordination named as the main blocker.
• · Roughly one in ten financial institutions has deployed compliance agents at scale despite vendors claiming one officer can supervise 15-20 agents.
• · NAB is building a central registry of agents, models, owners and risk ratings before wider rollout rather than expanding deployment first.
• · A Swiss pilot reached 99% document-extraction accuracy, indicating capability is not the limiting factor.

Counterfactual — If model capability were the constraint, the pilot-to-production gap would track model performance and would narrow as frontier models improve. It has not — and institutions running the same models report very different production rates, which points at what sits underneath them rather than at the models themselves.

Post-quantum migration became a live deadline through procurement and vendor contracts rather than through prudential regulation — but that is now changing, and the banking sector has responded faster than its own stated timelines predicted.

• · SWIFT is targeting a PQC-enabled SwiftNet by 2027 with a 15-month migration window for participating institutions.
• · All remaining FIPS 140-2 validated modules moved to Historical status on 21 September 2026, cutting them off for new US federal procurement.
• · The US Treasury's Quantum-Readiness Task Force, launched August 2026, has third-party and vendor readiness as a dedicated workstream.
• · An independent scan found banking's post-quantum TLS readiness rose from 28% in April to 94% by early September 2026.

Counterfactual — If this were regulator-led, the binding dates would be coming from prudential supervisors rather than from a messaging co-operative and a US federal procurement cut-off. That framing is only partly holding up: at least one prudential regulator has now issued a dated mandate, which is the evidence that would falsify the 'procurement not regulation' half of this claim.

Updated from the last snapshot. The read that Open USD threatened Circle's reserve-yield model was right about the economics and wrong about the casualty: reserve yield has indeed moved from the issuer toward distributors and holders, but Circle was not displaced — it became the infrastructure underneath that shift.

• · The previous snapshot carried this inference forward unchanged, with no fresh Circle or Open USD development to test it against.
• · Nu Global converts customer deposits into USDC and EURC and pays holders 3.5% and 2.2% APY respectively — yield reaching the holder, not the issuer.
• · Corpus reporting puts USDC liquidity at US$73.3bn with USDT holding around 60% of a roughly US$308bn market.
• · Mastercard bought stablecoin infrastructure firm BVNK outright for up to US$1.8bn rather than building or partnering.

Counterfactual — If Open USD had displaced Circle as the last snapshot implied, USDC circulation would have contracted and major distributors would be choosing OUSD for new products. Instead a top-tier neobank built its new US multicurrency account on USDC — which falsifies the displacement half of the claim while leaving the yield-migration half intact.