Wells Fargo puts AI agent losses on you, starting November 24

Share
Wells Fargo puts AI agent losses on you, starting November 24

On Tuesday, Wells Fargo's revised Online Access Agreement takes effect for millions of customers, and the new Section 15 is the reason to read it before you connect anything to your bank account. "Your use of any AI Agent is AT YOUR OWN RISK," the clause says, in the bank's own capitals, per the agreement text.

It is the bluntest version of a shift that ran through the largest US banks in about five weeks. JPMorgan Chase blocks third parties that log in with a customer's credentials. Bank of America told customers not to hand logins to agents. Citigroup added a Section H deeming agent activity the customer's own, and PNC changed its terms on November 18, TechCrunch reported. Charles Schwab's chief executive told Bloomberg the firm blocks unauthorized agents trying to move client money 10 to 20 times a week. Wired surveyed the group this weekend and found five banks reaching the same destination by two roads: Chase bans the login, Wells Fargo permits it and moves the loss.

What Section 15 actually does

The definitions do the work. An "AI Agent" is "any software, application, program, or other technology that takes autonomous, semi-autonomous, or non-autonomous action on behalf of any person or entity" — which reaches budgeting apps, browser extensions and homemade scripts.

Then the authorization clause. If you give an agent your login or otherwise authorize it, any instruction or transaction it initiates is "deemed to have been initiated and authorized directly by you, without further verification," and the bank has "no obligation to investigate the validity or scope of such authority."

Then the responsibility clause: you are "solely responsible" for monitoring the agent and keeping it inside the authority you granted, and — unless law says otherwise — for "unintended actions," for actions from "incomplete, inaccurate, or manipulated inputs," and for the agent exceeding what you allowed. The bank says it does not monitor, screen or review agent activity.

Then the clause a lawyer would circle. You must revoke authorization "properly and timely," retain evidence that you did, and produce it when the bank asks. Until you notify Wells Fargo, you "remain responsible for all the AI Agent's activities." Turning the agent off in its own app is not the same as telling the bank. And by using an agent you "expressly consent" to sharing financial and personal information with it, outside the bank's privacy policies.

A ban would be cleaner. This is a liability transfer written as a disclosure, and its effect is to turn a question about intent — did you mean to buy that? — into a question about authorization — did the software have your password?

Why the banks moved now

Two things happened in October. Banks began detecting third-party agents logging into customer accounts and, in some cases, acting beyond what the customer asked; executives said they struggled to tell which agents had been properly authorized, Decrypt reported during the scare. The agents also kept supplying the bad examples on their own.

The volume is still close to nothing. DoorDash said in October that third-party agent orders numbered in the several hundred against 993 million orders in the quarter, as we noted in DoorDash: several hundred agent orders against 993 million in Q3. Forrester's Sucharita Kodali told Wired that most consumers do not know what an agent is. Banks are writing for a future they expect.

They are also building the alternative while they write it. Wells Fargo's generative-AI assistant has passed a billion customer interactions, Bank of America reworked EricaAssist this year, and Citi built an internal platform for agents. Close the front door to outside agents, open a side door for your own.

The Regulation E problem

The fight is not about whether an agent can be trusted. It is about whether a contract clause can reclassify a transaction consumer protection law already classifies for you.

Regulation E caps a consumer's liability for an unauthorized electronic transfer at $50 if it is reported within two business days and $500 after that, and puts the investigation duty on the bank. The clause's escape hatch is "unless otherwise provided by law or regulation," which every one of these agreements carries. If a transfer your agent initiated counts as authorized, the caps may not apply at all — including when a prompt injection or a leak, rather than your instruction, produced it. Citi's version goes further: the customer stays responsible even if the tool leaks credentials and someone else uses them, per TechCrunch.

No court has tested this. Hunton Andrews Kurth's comparison of the four October moves flags authority and attribution as the open questions, and consumer advocates have the better statutory argument in the obvious case: a stolen password used by a stranger is not something the account holder authorized, whatever a click-through says. The risk is the ambiguous middle — an agent was authorized, went off-script, and the bank's system logs a valid login.

Who wins, who loses

Registered agents win. American Express launched a developer kit in April that verifies agents before they transact and extends purchase protection to purchases made by agents a card member has registered, per Amex's announcement. That is the model the industry is converging on: the agent gets an identity, the credential is a token scoped to a mandate, and the loss lands somewhere defined. It is the part of agentic commerce that AI 101 — What is agentic commerce? argued was genuinely hard — not the shopping, but proving who allowed the spend.

Credential-based agents lose, which means the personal-finance startups whose entire pitch was "we log in as you and do the work." Read-only access through open banking connectors such as Plaid still works at Chase; anything that moves money does not. Platforms have been here before — Resy banned diners whose AI agents grabbed the tables — and the pattern repeats: whoever owns the account relationship decides who may act, then sells access back on its own terms.

The counter-case

The banks are not wrong about the risk. Credential sharing is a major vector for account takeover, and an agent that reads web pages can be prompt-injected. A bank cannot audit software it did not build, and it still answers to regulators for fraud. Writing a clause that says so is defensible; writing one that also disclaims any duty to notice a pattern of agent-driven withdrawals is a different thing.

There is also a limit to what the terms settle. They qualify themselves with "to the extent permitted by law," an admission that the statute outranks the contract. And Wired's experts put the threat in perspective: people are comfortable letting an agent buy toilet paper, less comfortable letting it refinance anything. The disaster here is ten roast chickens, not an emptied retirement account.

What to watch

Whether a state attorney general or the CFPB tests Section 15 against Regulation E. Whether Wells Fargo's reserved right to require agent registration becomes a real registry with Amex-style protection attached. And whether Chase's updated Digital Services Agreement, effective January 1, does what its block does not: give customers a lawful path to a first-party agent while the third-party door stays shut.

Until then, the practical reading is narrow and unpleasant. If you connect an agent to a bank account, assume every action it takes counts as your decision, assume you will need to prove you revoked access, and assume the fraud protections you paid for were written for a world where the only actor with your password was you.

Does your bank let an agent touch your account — and did you read what you agreed to? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Resy banned diners whose AI agents grabbed the tables

Resy banned diners whose AI agents grabbed the tables

Reservations became the first consumer service where delegating to your AI assistant is a terms-of-service violation — and the diners getting banned are the ones the platform is quietly building agent access for. Meanwhile, China's biggest AI image company shipped a studio built for the same one-person crew. Resy has been deactivating the accounts of diners who pointed personal AI agents at its booking system, and it will not say how many.

Warcraft Logs is tightening its API as AI addons flood WoW: Forever

Warcraft Logs is tightening its API as AI addons flood WoW: Forever

World of Warcraft: Forever goes live at 23:00 UTC on November 4, Blizzard says — about 32 hours after this is published. It arrives as a third permanent version of Azeroth, with more than 1,000 new quests and its first level-60 raids on December 9, per Blizzard's launch announcement. It also arrives with a problem no WoW launch has had before: the addon ecosystem around it is already full of AI-generated clones of other people's work. The flood arrived before the servers Joseph Cox…

Google ties Gmail and Photos data to ads via Personal Intelligence

Google ties Gmail and Photos data to ads via Personal Intelligence

Google's advertising section has said the same thing for five years: personalized ads are not built from the contents of your Drive, Gmail or Photos. As of October 16, that sentence comes with an exception — and the exception is the AI feature the company spent this year persuading people to switch on. One clause, four edits The old bullet read: "We don't show you personalized ads based on your content from Drive, Gmail, or Photos." The new one keeps every word and then continues …

AI 101 — What is AI memory?

AI 101 — What is AI memory?

AI memory is a feature of the app, not of the model: it stores notes about you somewhere outside the model and feeds the relevant ones back into the prompt on later turns, so a new conversation can start with some of the old one still attached. The model underneath remembers nothing. Its weights are frozen after training, and between requests it holds no state at all. That split is the whole concept, and almost every confusion about AI memory comes from missing it.

Anthropic cut junior hires to 9% — who trains the experts now?

Anthropic cut junior hires to 9% — who trains the experts now?

Anthropic has published the first frontier-lab account of what AI agents do to the people who work beneath them, and the headline is that the lab is stopping halfway up the career ladder. The economics team's report, How employment is changing at Anthropic, is built from internal hiring records, agent telemetry, employee surveys, calendar data and 17 staff interviews. We walked through the core numbers yesterday — Anthropic's junior tech hiring fell to 9% as agents outworked staff.