A global Microsoft 365 outage disrupted financial firms’ operations on 13 November, delaying communications and prompting one Taiwanese firm to cancel a scheduled bond repurchase transaction. Microsoft said there was no indication of a cyberattack as it worked to stabilise affected servers.

Nubank linked slower hiring to increased staff capacity, while ANZ reported that AI had doubled coding capacity on its Australian banking project. At Sberbank, 55% of new lending transactions for medium-sized and large companies now run from application to disbursement without manual intervention.

Read more on the week’s key developments:

1. Microsoft says no sign of cyberattack as outage disrupts financial firms

A global Microsoft 365 outage disrupted financial firms’ operations into Asian business hours on 13 November. In Taiwan, one financial firm cancelled a scheduled bond repurchase agreement transaction, Reuters reported, citing a person familiar with the situation. As email services faltered, banks redirected customer enquiries and urgent corporate banking requests to telephone support and other channels. The disruption also reached Microsoft connectors used by OpenAI and Anthropic.

Efforts to restore service repeatedly stalled. Microsoft said unexpectedly unavailable servers had strained its remaining infrastructure and identified an overwhelmed process handling secure connections. Some servers stopped responding again when traffic returned, preventing adjustments from delivering sustained recovery. Email, calendars and connected applications were affected through their dependence on Exchange Online. Microsoft said it was not aware of malicious activity causing the outage; its investigation into the underlying cause continued.

2. Nubank slows hiring as AI expands staff capacity

Brazil’s Nubank said on 12 November that it expected its 2026 efficiency ratio to reach 19%–19.5%, improving on its previous estimate of around 20%. Chief financial officer Rob Livingston told Reuters that AI had increased staff capacity and management had hired more slowly than planned. During the third quarter, the group added 3.7 million customers to reach 143 million, while its efficiency ratio fell from 19.5% to 19.3%.

Customer-service testing offers one example of the operational gains. On 5 November, Nubank reported that testing AI agents on simulated customers had shortened the average development cycle for an updated version from 21.2 days to 4.4 days. Stronger revenue also contributed to the efficiency-ratio improvement, and some expenses shifted into the fourth quarter. The bank has not quantified recruitment savings attributable to AI.

3. Sberbank automates 55% of new loans to medium-sized and large firms

Russia’s Sberbank said on 12 November that 55% of new lending transactions for medium-sized and large companies now run from application to disbursement without manual intervention. First deputy chairman Alexander Vedyakhin separately put the value of loans issued with AI assistance to medium-sized, large and very large businesses at about RUB 8 trillion ($96.8 billion). How much of that sum went through the fully automated process was not disclosed.

The expansion comes amid weakening corporate credit quality. At the same investor day, chief executive German Gref reported a sharp deterioration in the second half of 2026. Sberbank provided no performance breakdown by approval method, leaving unclear how arrears and losses on fully automated loans compare with those reviewed by staff.

4. Australia puts AI disruption into its 2027 financial system stress test

AI-driven economic disruption will feature in Australia’s 2027 financial system stress test, the Australian Prudential Regulation Authority (APRA) said on 12 November. The exercise will cover banks, superannuation funds and life insurers. One scenario considers how rapid adoption could undermine borrowers’ business models; the other examines a sharp fall in the value of AI-related assets. APRA’s previous system-wide exercise examined links between banks and superannuation funds.

For banks, the losses could emerge through borrowers whose revenues and debt-servicing capacity weaken. Pension funds and insurers could suffer valuation losses before defaults occur. By testing the sectors together, APRA can examine how the same shock affects interconnected institutions, including those that make little use of AI themselves.

5. ANZ doubles coding capacity on its Australian banking project

ANZ chief executive Nuno Matos said on 9 November that AI-assisted engineering had doubled the team’s capacity to produce code for the digital interface it is building for Australian retail and small-business customers. At the end of September, the project was 45% complete. ANZ aims to reach 80% by March 2027 and deliver the interface to 7.7 million retail customers and more than 600,000 small businesses by the end of September 2027.

Using AI for routine code allows engineers to work on several tasks in parallel, while they retain responsibility for design and quality checks. The completion percentage tracks approved design decisions and work packages, rather than code produced. ANZ is also introducing controls to trace agents’ actions and stop them when they exceed their authority.

6. Northern Trust cuts investment data preparation from weeks to three days

An AI agent has cut average data preparation time from two and a half weeks to three days in live institutional onboarding cases, Northern Trust and Microsoft reported on 9 November. The US bank uses the agent in Front Office Solutions, its reporting platform for institutional investors, to convert years of transactions from client spreadsheets into a standard format. The timing includes analyst review, exception handling and validation.

Reconstructing investment histories is complicated by fund managers using different labels for the same transaction. In a separate test on seven anonymised workbooks, the agent achieved a first-pass F1 score, which combines precision and recall, of 96.5%. Analysts still reconcile records with the outgoing administrator before a client goes live. The three-day figure therefore covers preparation and review, rather than the entire onboarding process.

7. ECB assesses banks’ plans to address frontier-AI cyber risks

The European Central Bank (ECB) said on 9 November that it was assessing banks’ action plans to address frontier-AI risks, following an end-October submission deadline. In a written overview for the Eurogroup, supervisory board chair Claudia Buch warned that increasingly capable models were making it quicker and cheaper to identify and exploit IT vulnerabilities. The assessment will identify areas requiring improvement and inform feedback on good practices across the industry.

A July supervisory letter on AI-enabled cybersecurity threats preceded the exercise. Figures cited in the overview show that more than 90% of ECB-supervised banks use AI tools and 85% use generative AI. Alongside the threat from attackers, the ECB highlighted banks’ dependence on external digital providers and their ability to maintain critical services during disruptions.

8. South Korea closes financial assessment portal after data breach

South Korea’s Financial Security Institute blocked access to its assessment portal after detecting unauthorised access on 9 November. Personal details in forms submitted by financial firms and about 350 portal-user records were exposed, the institute confirmed. Yonhap reported on 12 November that roughly 130 files from 48 financial firms were estimated to be affected, with the Financial Services Commission due to begin a targeted audit on 16 November.

Although the institute masked names and mobile numbers in user profiles, uploaded forms contained details including names and dates of birth. Genuine assessment documents and contact information could support impersonation attempts, such as fraudulent requests for firms to resubmit sensitive files.

9. SBI Shinsei extends voice AI to a line handling 250,000 enquiries a year

Japan’s SBI Shinsei Bank expanded its voice AI service on 11 November from specialist telephone lines to an enquiry desk handling about 250,000 calls annually. Callers describe their request in their own words, and the system answers routine questions or refers cases requiring identity checks or individual action to employees. Customers can also request a person. The bank aims to resolve 40% of enquiries through AI.

That target would represent about 100,000 enquiries a year at current volumes. Before launch, the bank recorded more than 97% correct final answers across roughly 6,200 test conversations on common questions.

10. North Pacific Bank reports 70% AI adoption among sales staff

Japan’s North Pacific Bank reported through technology partner SoftBank on 10 November that 70% of its sales staff used its generative AI service. Since rolling out the service in September 2025, the bank has provided about 2,500 staff accounts. Branch training and prompts for specific tasks accompanied a tripling of users and a 7.1-fold increase in usage over a year. Some employees estimated that entering customer meeting notes took about 40% less time.

An early attempt to draft lending recommendations exposed a gap in the underlying records: staff sometimes entered too little detail after customer meetings for the AI to produce a useful credit memo. The bank changed how it captures conversations, using AI to summarise them for the customer record. It plans to test AI-assisted analysis of borrowers’ finances as a separate application.