JPMorgan keeps outside agents out, Wells Fargo and DBS use AI inside

JPMorgan keeps outside agents out

JPMorgan Chase CEO Jamie Dimon said the bank won't allow outside AI agents to access customer accounts and operate on their behalf, citing risks to the company and its customers. A spokeswoman confirmed the bank currently blocks external agents from operating on behalf of retail customers, with no timeline for safeguards that would let them interact directly with accounts.

Read-only access still works

Agents can still access some customer data in a read-only fashion, such as balances, if Plaid or another connector has the user's credentials, CNBC reports. They can't send money or pay a bill. Dimon added that people using them should understand there is a liability transfer.

Banks already put AI to work

Wells Fargo is using AI and, in some cases, autonomous agents to write code faster, launch marketing campaigns faster, and accelerate response times for underwriting decisions and customer complaints, PYMNTS reports. Scharf said the bank is taking a disciplined approach to managing risk. DBS, for its part, says it runs more than 430 AI use cases, and its 2025 annual report estimates about S$1 billion in economic value from its data analytics and AI and machine-learning initiatives.

At DBS, humans still sign off

In August, DBS gave around 1,500 relationship and credit risk managers worldwide a tool that drafts a corporate credit memo, the document that makes the case for lending to a company. The tool drafts the memo, and bankers and credit risk managers still make the assessment and approval decisions.

Someone has to measure agent risk

Harbor came out of stealth with $10 million in seed funding and says it measures the operational, security and regulatory risk of each AI agent a company runs, with banks and insurers in the United States, Europe and Latin America as its first customers. Insurance Journal reports that insurers do not have the claims history to price the risk of AI agents. If you work in a risk or operations team, a useful question to ask this week is which agents already run inside your own walls, and who signs off on each one.