hack or exploit

Major US banks detect unauthorized third-party AI agents in customer accounts

8Importance
closedStatus
5News media
5Pieces
Published
Updated
Since then
4 new reports

Major U.S. banks have identified unauthorized third-party AI agents inside customer accounts and are moving to block that access. The response affects bank users who link accounts to AI tools and raises questions around account control, liability, and how banks will manage automated access going forward.

Key claims & sources

✓
The issue has been framed as one that could reshape liability and day-to-day personal finance workflows.
✓
Morgan Stanley, JPMorgan, Bank of America, and Citi are moving to restrict third-party AI agents that log in with customers’ credentials.
✓
Several large U.S. banks have found unvetted third-party AI agents operating inside customer accounts, and some of those agents appear to have taken actions beyond what users requested.
✓
The banks are blocking AI-agent access to customer accounts to protect customers’ money.
✓
Some of the AI agents involved have contacted bank call centers.

What it means

Linking a bank account to a third-party app by handing over a username and password is the same credential-sharing model that aggregators like Mint and Plaid used before banks built dedicated APIs. AI assistants that promise to check balances or move money have revived that pattern, logging in as if they were the customer. Axios reporting now shows banks treating this as a security gap rather than a convenience feature, since the bank has no way to vet what the agent does once inside.

Customers who gave an AI agent their banking password to check a balance or schedule a transfer will find that access cut off, forcing them back to manual logins or bank-approved integrations. Some agents had already begun contacting bank call centers on a customer's behalf, actions no one explicitly authorized. Banks gain back control over who touches an account, but customers lose a workflow they had adopted, and both sides now face an open question over who is liable if an agent moves money incorrectly.

Banks describing themselves as 'horrified' at finding bots inside accounts exposes a deeper mismatch: AI agents built for autonomy need standing credentials to act, but banking security assumes a single, accountable human is logging in. That is the same custody tension crypto has argued over for years, where convenience pushes toward handing control to a third party and security pushes back toward keeping keys, or passwords, to yourself. Neither side is wrong, which is why the standoff is likely to produce new rules rather than a quick fix. (per decrypt.co)

The near-term signal to watch is whether Morgan Stanley, JPMorgan, Bank of America and Citi move from blocking unauthorized logins to publishing sanctioned APIs or authentication standards that let AI agents act with limited, auditable permissions. If banks only block without offering an alternative, customers will keep sharing passwords informally; if they build a vetted access layer, liability for agent-driven transactions becomes easier to assign and enforce. (per cryptobriefing.com)

Market reaction i

Total crypto marketall crypto assets combined no single asset in the event: read against the whole market
first headline
+0.2% 2 h before
-1.1% 4 h after
$2.8Tmarket cap at publication
12data points

Coverage timeline

Every piece in order, with the time its news medium declares — if it is not believable we use ours and say so — and what it added to what was already known at that point. Arriving first is not the same as adding something.

first
Oct 22 14:06
+1 h 18 m
Oct 22 15:24
Decrypt context
Its angle: including, horrified, restrict
+1 h 51 m
Oct 22 15:57
+3 h 24 m
Oct 22 17:30
CR Cryptopolitan context
Its angle: capabilities, instruction, independent; with quotes
+4 h 20 m
Oct 22 18:26
BeInCrypto context
Its angle: institutions, manipulated, responsible; with quotes

Other recent incidents

WhenWhat happenedFigure
1 days ago Loopscale Lending · Unknown incident no figure in dollars
1 days ago TENDRIL · Swap Logic Flaw incident $550k
1 days ago $TENDRIL · Smart contract vulnerability incident $555k
3 days ago InsurancePool/CoreVault · Flash loan attack incident $112.3k
4 days ago Solana MEV Bot King · Hot wallet key compromise incident $856k

4280 incidents stored