More banks says customers using AI agents are on their own — even if they get hacked

Glass Epoch DeskOriginal digestSource: TechCrunch AI

AI-assisted digest, checked against the original source before publishing. How we work

Citibank joins JPMorgan, Bank of America, PNC and Citizens in making customers absorb losses from third-party AI agents — including hacks of the tools themselves.
More banks says customers using AI agents are on their own — even if they get hacked
Photo: Jonathan Cooper · free stock license

Citibank updated its online user agreement this week to tell customers who connect third-party personal finance AI tools to their accounts that the fallout is theirs to absorb. The bank said it wants customers to understand the risks of sharing banking credentials or authorizing outside agents, and that anyone who grants an agent access is responsible for what that agent does — intentional transactions, mistakes, or both.

The policy shift is spreading quickly across the sector. JPMorgan and Bank of America announced comparable terms last month, and PNC revised its own agreement the same day as Citi's disclosure, making this a coordinated posture among major US lenders rather than a single bank's caution.

Stated security risk is only half the motivation. Banks also treat agent-driven personal finance as an encroachment on their franchise: markets briefly reacted to the prospect of AI tools steering consumers away from unnecessary fees and low-interest deposit accounts that pad bank margins. Citi's terms draw a telling contrast — AI agents it provides itself can be used without worrying about additional liability, while outside agents carry the burden. Anthropic's own approach underscores the caution: the personal finance tool it launched this month is billed as read-only and cannot move, manage or invest money.

For AI startups, personal finance remains an attractive route to new revenue and a share of the sector's fee stream, but bank accounts present a far stiffer trust test than email or chat history. What happens next hinges on open questions: whether banks keep tightening liability terms as agents grow more capable, whether consumers accept that trade-off, and whether read-only designs become the industry default. As with car-hailing and food service before it, how service providers negotiate control of the agent pathway will decide which AI investments actually pay off.

Key facts

  • • Citi ranks as the third largest commercial bank in the United States.
  • • JPMorgan and Bank of America, the two largest US banks, moved last month.
  • • PNC, the eighth largest US bank, changed its terms the same day.
  • • Anthropic's personal finance tool launched this month as strictly read-only.
  • • Citi's terms shield liability for AI agents the bank itself provides.

Why it matters

Builders shipping agents with financial write access now face a market where major banks push fraud and hack losses onto the account holder, raising the cost of disputes and refunds. Expect more vendors to default to read-only designs and to negotiate bank partnerships instead of credential-sharing, while liability language becomes a standard diligence item for agent startups.

Spotted an error? Send a correction.

Get the day's AI news in one email

The Daily Epoch · 60-second read · unsubscribe anytime

Related