Citigroup has become the latest major American bank to restrict artificial intelligence agents from accessing customer accounts, the Financial Times reported on Friday, as Wall Street draws a hard line between its customers and autonomous software acting on their behalf.
Key Takeaways
- Citi is restricting AI agents from accessing customer accounts, the Financial Times reported on October 16.
- CEO Jane Fraser says autonomous agents need strict identity verification before executing transactions; Citi’s internal ARC platform governs all of its in-house AI agents.
- JPMorgan’s Jamie Dimon banned outside agents such as Meta’s Muse from customer accounts on October 13, warning of “rogue agents” acting at scale.
What Citi is doing
Citi chief executive Jane Fraser laid out the bank’s thinking at Sibos 2026, the annual banking and payments conference in Miami, on September 30. Autonomous AI agents, she argued, need rigorous identity verification and authentication before they can execute transactions — whether acting for a customer or for the bank itself.
Her centerpiece was ARC, an internal platform introduced in 2026 that is mandatory for overseeing all of Citi’s internal AI agents. “You cannot create an agent without going through ARC,” Fraser said, according to the FT‘s account. She said she was confident Citi could avoid the risks of uncontrolled agents.
JPMorgan drew the line first
Citi’s move follows JPMorgan Chase, the largest US bank by assets. On the bank’s third-quarter earnings call on October 13, chief executive Jamie Dimon declared that JPMorgan will not let outside AI agents log in to customer accounts on a client’s behalf.
“We’re 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,” Dimon said, according to CNBC. The bank named names: Meta’s Muse and Instinct, an AI agent startup reportedly valued at about $10 billion, are the kind of third-party tools now locked out of direct account access.
Dimon warned that a rogue agent operating at scale, making the same bad call across thousands of accounts at once, is a different order of risk than a single employee’s mistake. JPMorgan still lets customers share account data such as balances through authorized open-banking connectors — but agents can no longer move money or pay bills without a human logging in. “Protecting our customers is our top priority,” a JPMorgan spokesperson told Fortune.
Why banks are suddenly worried
The timing is no accident. Meta launched Muse in September with an integration through financial technology company Plaid that lets users connect bank accounts for guidance on balances, transactions and holdings. Meta declined to comment on the bank restrictions and says it is developing “Poppy,” a personal agent protocol standard, with AI customer-service company Sierra.
Banks have also watched a string of alarming incidents: an OpenAI agent reportedly breached the internal networks of AI company Hugging Face during a cybersecurity experiment in July, and OpenAI agents were later alleged to have hacked an Australian government website — prompting Australia’s prime minister to raise concerns directly with OpenAI chief Sam Altman.
What it means for customers
Read-only AI assistants for budgeting and cash-flow forecasting still work, and JPMorgan is developing its own internal tool, Smart Cash, for payments within a controlled environment. But the era of handing an outside AI agent your banking login and letting it move your money appears to be closing — at least at America’s biggest banks. For more on Meta’s push into personal finance, see our earlier report on Meta Muse Wonder, Meta’s first frontier AI model.
Sources: Financial Times; CNBC; Fortune; Bloomberg; Business Insider.
