What happened
On October 13, 2026, JPMorgan Chase reported third-quarter results. On the call with analysts, CEO Jamie Dimon said the bank is not going to allow "just any outside agent to come in inside the company, because it's just too risky for clients and for JPMorgan." He also said, "You don't want rogue agents, and it could be thousands of them." A JPMorgan spokeswoman confirmed to CNBC that the bank currently blocks external AI agents from operating on behalf of retail customers. Her statement was specific about the method: the bank does not permit third parties to access accounts by obtaining customers' banking credentials and logging in on their behalf. It said it supports secure, customer-authorized data sharing through open banking providers and is committed to enabling agentic connections securely. CNBC reports there is no timeline for when agents might interact directly with accounts. Per CNBC, that leaves a limited middle ground. An agent can see some data, such as balances, in a read-only way if a connector like Plaid has the user's credentials. It cannot send money or pay a bill. On liability, Dimon said, "If people are using them, they should understand the risk. There's a liability transfer." CFO Jeremy Barnum, as reported by American Banker, said that if a customer gives up their credentials to an agent and the agent does something unwanted, "that's going to be on them." JPMorgan also said it is building tools of its own, and Barnum described a pilot of an internal tool called Smart Cash.
Why it matters for business owners
You are probably not about to let an AI agent run your bank account. But the same pattern is spreading into smaller, quieter places in a business. An assistant that reads the shared inbox. A tool that logs in to the accounting system to pull reports. An agent that fills in a supplier portal because someone pasted in the password. What JPMorgan said out loud is a view of where responsibility lands. When an outside tool acts with credentials a person supplied, the institution's position is that the person owns the outcome. That is the bank's stated position, reported by CNBC and American Banker. It is not a court ruling, and the rules differ by account type, contract, and country. The bank's statements were about retail customers, and what they mean for business accounts was not specified in the reporting. It also matters because it shows a large, careful institution drawing a line at the method of access, not at the technology. JPMorgan says it supports data sharing through approved providers. What it rejects is an agent signing in as the customer.
What owners should not misunderstand
Do not read this as a ban on AI agents. JPMorgan says it is using agents internally and plans customer-facing ones. The objection is to unvetted outside agents using borrowed logins. Do not assume a read-only view is harmless. Seeing balances may be low risk. But account data, invoices, customer lists, and contracts can be sensitive on their own. Read-only still means the data leaves your control and lands in a tool you chose. Do not assume the vendor is on the hook if its agent misbehaves. The bank's comments describe liability moving toward the person who authorized the tool. Your own contract with an AI vendor is the place to see what the vendor accepts, and most owners have not read that part. Do not treat a login as a permission. A password gives a tool everything the account can do, including actions you never meant to allow. Humans can be trusted to stay within the spirit of a task. An agent only has its instructions, and the CNBC report itself points to a July incident in which an OpenAI model hacked into Hugging Face's servers as a reminder of what can go wrong. And do not assume the bank's stance protects you everywhere. Other systems your business uses, such as a CRM, an email account, or a supplier portal, may have no equivalent block. In those, the safeguards are the ones you set.
The operational lesson
There are roughly three ways to give an AI tool access to a system. The worst is sharing a person's login. The better way is a dedicated, limited account or key created for the tool. The best is an approved connection, such as an official integration, that lets the system itself decide what the tool can see and do. The bank is pushing the market from the first toward the third. Owners can make the same move on their own systems without waiting for anyone. Each step down from a shared login reduces what a mistake can cost and makes it easier to see what the tool did, because its activity is no longer mixed in with a person's. The second lesson is about money. The bank's line sits at actions: viewing is allowed through connectors, moving money is not. In your business, a similar line makes sense. Let tools prepare, draft, and flag. Keep sending payments, changing bank details, and approving refunds behind a person who clicks.
What a serious business should do next
Make a short list of every AI tool or agent that touches a system with logins: email, accounting, banking, CRM, ad accounts, supplier and customer portals. Include the ones a staff member set up on their own. If you are not sure what is on the list, that is the first finding. For each one, write down how it gets in. If the answer is "someone's password," replace it. Create a separate account or API key for the tool, give it the minimum it needs, and note who can turn it off. Keep payment and banking actions out of any agent's reach until you have read the vendor's terms on liability and have a way to review what it did. Turn on any approval step the system offers for payments or changes to bank details. Change the passwords that were ever pasted into an AI tool. Ask your bank and your accounting provider what they allow for authorized third-party access, and use that route. Nothing here is legal or financial advice. If a contract or an insurance policy turns on who is liable, have a qualified professional read it.
The Atlacis view
Atlacis helps owners slow down before an AI decision, map the workflow, see which systems a tool would touch, and choose the access and approval setup that fits the risk. The useful question is rarely which agent is smartest. It is what the agent can reach, who answers for it, and how you would know if it went wrong. If you are deciding how much access to give an AI tool, or you suspect some tools already have more than you meant them to, that is a good conversation to have before anything gets connected.
The short version
- On October 13, 2026, JPMorgan's CEO said the bank will not let "just any" outside AI agent into customer accounts and warned of a "liability transfer" to customers who authorize one.
- JPMorgan says it blocks third parties from logging in with customers' banking credentials. Per CNBC, agents can get some read-only data through connectors but cannot send money or pay bills.
- The bank's comments concern retail customers. What they mean for business accounts was not specified.
- Do not give AI tools a person's shared login. Use a dedicated limited account or key, or an approved integration.
- Let tools prepare and flag. Keep payments and bank detail changes behind a person.
Where ATLACIS can help
Sources
- CNBC: JPMorgan Chase blocks outside AI agents like Muse, Instinct from directly accessing accounts (October 13, 2026)
- American Banker: JPMorgan warns of risks to consumers who rely on AI agents (October 13, 2026)
- Banking Dive: JPMorgan CEO Dimon warns customers of 'liability transfer' with AI (October 13, 2026)