AI in financial services this week

Week ending 19 October 2026 · 41 sources

Banks shut the door on outside AI agents as supervisors tighten focus

JPMorgan, Citi and Schwab restricted third party AI agents this week, and G7 economies planned simulated AI attacks on banks 1234. The FSB finalised 12 sound practices for AI adoption by financial institutions 5.

Which way the market is moving

US banks lock outside AI agents out of accounts

JPMorgan, Citi and Schwab restricted outside AI agents from customer accounts this week 123.

JPMorgan blocked outside AI agents from directly accessing customer accounts, and Citi became the latest US bank to restrict them 12. Bloomberg reported US banks blocking agents that attempt to move money for customers, and Schwab blocked unauthorised agents while pointing clients to its own assistant, Charley 163. Jamie Dimon also warned people not to let AI manage their money 17.

The pressure is not going away. M&T said it is monitoring the impact of AI agents on deposits, and Curinos says AI rate shopping is already here 1819. FIS says banks must verify which AI agent initiated a transaction, and some banks are exploring bring your own agent models 76.

What it means: Any institution with online accounts now needs a stated policy on how outside agents are identified, permitted or refused. Deposit and savings products may also face more automated rate comparison.

Questions for vendors:

  1. How does your platform tell an AI agent session apart from a customer session?
  2. Can you record which agent initiated a transaction and under whose authority?
  3. What scoped, revocable permissions do you support for customer-authorised agents?

In the registry: Tanium Atlas (Tanium), AgentNext (BUSINESSNEXT), Insurer Solution (Acturis)

Supervisors focus on AI-enabled cyber attacks

G7 economies plan simulated AI attacks on banks and the Bank of England urged banks to prepare for AI cyberattacks 49.

G7 economies plan simulated AI attacks on banks as cyber fears rise, and Insurance Business reported regulators sounding a systemic alarm 420. The Bank of England urged banks to prepare for AI cyberattacks 921. In New Zealand, the RBNZ and FMA set out supervisory expectations on cyber risks associated with frontier AI 10.

The work is already concrete. Bank of America brought in 1000 contractors to support 3000 internal staff patching software flaws uncovered by Anthropic's Mythos model 22. In Korea, AI-driven cyberattacks hit 16 financial firms and exposed 100,000 records 23. US and European regulators are also stepping up scrutiny of AI safeguards, Chinese models and cybersecurity 24.

What it means: Expect supervisors to ask how quickly you can find and fix flaws that AI tools surface, and how you recover from AI-enabled attacks. Patch capacity may need to scale well beyond normal staffing.

Questions for vendors:

  1. Have you scanned your code with frontier AI tools, and what did you fix?
  2. How quickly can you ship patches when many vulnerabilities are disclosed at once?
  3. Will you take part in our exercises that simulate AI-enabled attacks?

In the registry: AgentNext (BUSINESSNEXT), OCTO (TCG Process), AI Governance (ProductCloud)

FSB sound practices raise the accountability bar

The FSB finalised 12 sound practices to guide AI adoption by banks and other financial institutions 5.

The Financial Stability Board finalised a list of 12 sound practices to guide AI adoption by banks and other financial institutions 5. PYMNTS reported the FSB wants company-wide AI oversight inside banks, and Forbes framed the regulator view as trust but verify 2526.

CUBE says regulators are focusing on AI provenance and second-layer risk, so compliance teams can no longer rely on simply knowing that a rule exists 27. A report found banks face liability over their use of AI chatbots 28. Goldman Sachs chief David Solomon urged rigorous testing of AI and clear human accountability 29.

What it means: Boards and risk teams should expect AI oversight to be judged firm-wide, with evidence of testing, provenance and named human owners. Customer-facing chatbots carry legal exposure, not only service risk.

Questions for vendors:

  1. Which of the FSB's 12 sound practices does your product help us evidence, and how?
  2. Can you show the provenance of the data and models behind each AI output?
  3. What testing results and human review records can you hand to our auditors?

In the registry: AgentNext (BUSINESSNEXT), Insurer Solution (Acturis), IntelliDoc (ReconArt)

Institutions bring AI onto their own compute

Deutsche Bank, Starling and SBS moved AI onto on-premise or sovereign infrastructure this week 121330.

Deutsche Bank implemented an on-premise AI platform across its capital markets and investment banking teams in Europe and North America 12. Starling is working with Nvidia to build a transactional foundational model trained and hosted within a UK sovereign cloud 13. SBS launched a sovereign AI platform for core banking with NVIDIA 3031.

Cost and dependence are part of the reason. Farsight partnered with Ornn to help banks run AI on their own compute and cut reliance on AI giants 1432. Insurtechs are concerned about token bills that keep climbing, and an Insurance Journal viewpoint argues model independence matters for insurers 1533.

What it means: Data residency, model portability and usage cost are becoming design choices for AI programmes, not afterthoughts. Institutions locked to one model provider may find both cost and supervisory questions harder to answer.

Questions for vendors:

  1. Can your AI run on our own infrastructure or in a sovereign cloud?
  2. How easily can we swap the underlying model without rebuilding workflows?
  3. How do you forecast and cap token or compute costs as usage grows?

In the registry: RadiantOne Platform (Radiant Logic), Vibe (Mistral AI), Loan Origination (timveroOS)

Vendor agents move into compliance and engineering

FIS opened its AML investigations agent to non-clients and EXANTE reported results from an autonomous AI developer 3435.

FIS made its AI-based financial crime agent available to US banks and opened its AML agent to non-clients as alert volumes soar 3634. Customers Bank is using Palantir in a bid to become AI-native 37. Extend shared an AI receipt agent with bank partners for their small business customers 38.

Human control remains part of the design. EXANTE spent six months embedding an autonomous AI developer, Codey, into its engineering teams, reaching 75% success while humans kept control 35. Its experience suggests the main barrier to AI-led software delivery is organisational rather than model capability 35.

What it means: Agents for alert triage and software delivery are now packaged products, so buyers need clear lines on what the agent decides and what a person signs off. Organisational readiness may limit value more than the model does.

Questions for vendors:

  1. Which steps does your agent complete alone, and which need human approval?
  2. How do you measure agent accuracy, and can we see results on our own data?
  3. What audit trail does the agent leave for each case or code change?

In the registry: Tanium Atlas (Tanium), IntelliDoc (ReconArt), Tungsten TotalAgility (Tungsten Automation)

Regulation and supervision

Sources

  1. Finextra Research, 2026-10-15: JPMorgan Chase blocks outside AI agents from directly accessing customer accounts
  2. Financial Times, 2026-10-16: Citi becomes latest US bank to restrict AI agents from accessing accounts
  3. American Banker, 2026-10-15: Schwab blocks unauthorized AI agents, points clients to 'Charley'
  4. Financial Times, 2026-10-16: G7 economies plan simulated AI attacks on banks as cyber fears rise
  5. ABA Banking Journal, 2026-10-15: Financial Stability Board finalizes ‘sound practices’ for AI adoption
  6. American Banker, 2026-10-15: Why some banks are exploring 'bring your own AI agent'
  7. Yahoo Finance, 2026-10-16: FIS Global Says Banks Must Verify Which AI Agent Initiated the Transaction
  8. Startup Fortune, 2026-10-16: Self Labs Launches Agent Vault as Banks Lock AI Agents Out of Accounts
  9. ABC News - Breaking News, Latest News and Videos, 2026-10-16: Video Bank of England urges banks to prepare for AI cyberattacks
  10. Finextra (security), 2026-10-15: New Zealand regulators set out cyber and AI risk expectations
  11. The Japan Times, 2026-10-15: Banking watchdogs scrutinize AI use amid stability fears, sources say
  12. Finextra, 2026-10-16: Deutsche Bank rolls out on-premise AI system
  13. Finextra (retail banking), 2026-10-14: Starling becomes first bank to train AI model on UK sovereign cloud
  14. Yahoo Finance, 2026-10-15: Farsight Partners With Ornn to Help Banks Run AI on Their Own Compute
  15. Insurance Business, 2026-10-16: Rising AI costs concern insurtechs as token bills keep climbing
  16. Bloomberg, 2026-10-16: US Banks Block AI Agents Attempting to Move Money for Customers
  17. Yahoo Finance, 2026-10-14: Don’t let AI manage your money, warns JP Morgan boss
  18. American Banker, 2026-10-16: M&T says it's monitoring impact of AI agents on deposits
  19. Curinos, 2026-10-15: AI Rate Shopping Is Already Here – What It Means for Bank Deposits
  20. Insurance Business, 2026-10-16: G7 to war-game AI cyberattacks on banks as regulators sound systemic alarm
  21. PYMNTS.com, 2026-10-15: BoE Warns Banks Against AI Cybersecurity Risks
  22. Finextra, 2026-10-16: BofA called in 1000 contractors to patch flaws found by Mythos
  23. The Korea Times, 2026-10-14: AI-driven cyberattacks hit 16 financial firms, expose 100,000 records
  24. Yahoo Finance, 2026-10-15: AI Agents Could Pose New Risks to Banks: US, European Regulators Step Up Scrutiny of AI Safeguards, Chinese Models and Cybersecurity: Report
  25. PYMNTS.com, 2026-10-15: FSB Says Banks Need to Have Company-Wide AI Regulation
  26. Forbes, 2026-10-15: Trust, But Verify: How Regulators Want Banks To Use AI
  27. FinTech Global, 2026-10-16: CUBE: AI provenance and second-layer risk reshape compliance
  28. American Banker, 2026-10-15: Banks face liability over use of AI chatbots, report finds
  29. Banking Dive, 2026-10-15: What Wells Fargo, Citi and Goldman Sachs’ CEOs are saying about AI
  30. Yahoo Finance, 2026-10-14: SBS launches sovereign AI platform for core banking with NVIDIA
  31. The Paypers, 2026-10-14: SBS launches sovereign AI platform for banks with NVIDIA
  32. FinTech Global, 2026-10-16: Banks cut reliance on AI giants with Farsight and Ornn
  33. Insurance Journal, 2026-10-16: Viewpoint: Model Independence Is the AI Strategy Insurers Can’t Afford to Ignore
  34. FinTech Global, 2026-10-16: FIS opens AML agent to non-clients as alert volumes soar
  35. FinTech Global, 2026-10-16: EXANTE’s Codey hits 75% success, but humans keep control
  36. Yahoo Finance, 2026-10-15: Fidelity National Information Services Makes AI-Based Financial Crime Agent Available to US Banks
  37. FinTech Global, 2026-10-16: Customers Bank bets on Palantir to become AI-native
  38. Finextra (payments), 2026-10-15: Extend brings AI receipt agent to bank partners' SMB customers
  39. Financial Times, 2026-10-16: FirstFT: G7 plan simulated AI attacks on banks
  40. iTnews, 2026-10-15: In Pictures: Governing the agents of chaos Saviynt Melbourne roundtable
  41. AFR, 2026-10-16: Rise of the rogue AI bots a horror show for retail, banking and travel

About AI software

AI software across financial services: AI-native products and the machine learning, generative AI and agent capabilities inside banking, insurance and investment software.

Browse the products in the registry

Previous weeks

Get the market overview every Monday: subscribe to the Vederca newsletter.