Insights

The same category of technology banks are racing to deploy for efficiency — agentic AI — is simultaneously the vector for the newest generation of attacks against them. Deepfake scams and now a fully autonomous 'agentic' ransomware attack surfaced in the same window banks were announcing production agentic-AI rollouts, not in a separate news cycle.

Example: Researchers report first fully autonomous 'agentic' ransomware attack

NZ's move to single supervisors (DIA for AML/CFT, FMA for consumer credit) has now produced its first visible test case, and the bank on the wrong side of it is one of the country's largest — ANZ NZ, found in material breach of RBNZ outsourcing policy in the same edition the consolidation story otherwise looked complete.

Example: ANZ New Zealand found in material breach of RBNZ outsourcing policy

The productivity story and the displacement story are the same deployments viewed from two angles, not two separate trends. JPMorgan's onboarding cut from five days to under a minute and Westpac's six-day task cut to one hour are precisely the kind of gains now being cited as the reason banks are shrinking junior-analyst intakes.

Example: JPMorgan's agentic AI KYC push cuts onboarding from five days to under a minute

New this snapshot: the same AI wave driving new production deployments is forcing banks to admit their existing software is ageing faster than expected. NAB and BNZ's impairments aren't a rejection of AI investment — they're an accounting consequence of it, recognising that AI-era software has a shorter useful life than the assets were originally depreciated against.

Example: NAB follows subsidiary BNZ with A$1.3bn AI-driven software impairment

New this snapshot, and a clean three-edition arc: the 7 July edition had RBNZ's OCR call as a genuine knife-edge split. On 8 July, a Hormuz tanker strike sent oil back above $73, complicating the case for a hold. By 9 July, RBNZ delivered a surprise 25bp hike instead — a real-time example of a geopolitical shock arriving inside a live monetary-policy decision window, not just background noise.

Example: [UPDATE] RBNZ delivers surprise 25bp hike, lifts OCR to 2.50%

Agentic AI's productivity gains in banking are being captured primarily through workforce substitution rather than uniform redeployment — but this is a choice banks are making, not an inevitability of the technology.

• · JPMorgan and Westpac both report large efficiency gains (onboarding cut from 5 days to under a minute; a 6-day task cut to 1 hour) in the same window finance-sector job cuts were first reported.
• · By the 7 July edition, AI-attributed layoffs across finance and tech had reached 28,000/month and passed 100,000 for the year.

Counterfactual — If banks were mostly redeploying rather than substituting, we'd expect flat or growing junior headcount industry-wide alongside AI rollouts, not just at isolated banks who choose to hold the line.

Updated this snapshot — New Zealand's scam response is converging with Australia's on coordination, but deliberately not on liability. The cross-Tasman gap in who bears the loss looks durable, not a temporary lag Australia's model will eventually pull NZ toward.

• · NZ banks operate under voluntary scam reimbursement commitments (in force since the 20 June edition).
• · Australia's Scams Prevention Framework went live in the 7 July edition, making banks legally liable for prevention failures.
• · The NZ Anti-Scam Alliance's 2026 work programme (9 July edition) explicitly extends the existing voluntary banking/telco code model to more sectors, rather than introducing statutory liability.

Counterfactual — If NZ were actually converging toward Australia's model, the 2026 work programme would have proposed mandatory liability or penalties, rather than doubling down on voluntary sector codes.

Open USD's real threat to Circle isn't stablecoin competition on technology — it's an attack on the issuer-keeps-the-yield business model that has funded every major stablecoin issuer to date.

• · Visa, Mastercard, Stripe, Coinbase, BlackRock and 140+ firms launched Open USD in the 6 July edition.
• · Circle shares fell sharply in the days around the launch, reported in the 7 July edition.

Counterfactual — If this were wrong, Circle's stock reaction would track general crypto-market sentiment rather than move specifically on Open USD news, or Open USD's yield-sharing model would prove commercially unworkable once tested against real adoption.

New this snapshot — AI-driven software revaluation is becoming a recurring earnings-management lever for major banks, not a one-off accounting correction.

• · BNZ took a $253m post-tax profit hit ($352m software write-down) in its 2026 half-year results.
• · NAB followed with a $1.3bn underlying profit hit and $949m cash-earnings impact from the same policy change, reported in the 9 July edition.

Counterfactual — If this were a genuine one-off correction, NAB's leadership would frame it as an isolated true-up rather than part of a recognised pattern, and there would be no precedent of similar changes at the same institution.