Major US financial institutions have encountered a new category of operational risk — the unauthorized activity of AI agents in clients' personal accounts. In isolated episodes, autonomous bots made direct contact with bank call centers, forcing JPMorgan, Bank of America, Wells Fargo, and Morgan Stanley to accelerate the deployment of protective mechanisms.

No direct thefts from accounts in traditional banking have been recorded, but the crypto sector has already clearly demonstrated the danger of uncontrolled AI payments. In March 2025, an attacker hacked the AIXBT control panel and withdrew cryptocurrency worth $106,200. Two months later, a malicious message processed by the Grok neural network prompted the Bankr service to execute a token transaction worth $175,000. Ultimately, the Bankr team announced the return of 80% of the stolen assets.

Both incidents occurred in the blockchain environment and proved what consequences result from handing control of money to programs without an adequate level of security.

Liability for AI Agent Operations

JPMorgan CFO Jeremy Barnum addressed this issue during a corporate conference on third-quarter results on October 13, noting that clients who share credentials with agents that perform unwanted actions must understand that the responsibility lies with them.

US regulation limits consumer liability for unauthorized electronic transfers if they fall under the law. However, the client bears responsibility if they themselves granted the right to transfer and the executor exceeded that permission. It remains unclear how existing rules will apply to autonomous AI agents. It is precisely this legal uncertainty that will determine whether a financial institution is obligated to compensate for damages.

Why Major US Banks Are Blocking AI Agents

Of the 19 AI services and financial scenarios studied, 14 are capable of operating strictly within user-defined boundaries. In addition, five of the seven trading platforms analyzed directly prohibit agents from withdrawing capital. Banks are introducing similar restrictions.

Morgan Stanley has begun blocking third-party agents, while JPMorgan, Bank of America, and Citi are restricting their access. Morgan Stanley claims it has not identified unauthorized agents on its platforms. Some banks are switching to a read-only mode, giving agents access to financial information without the ability to transfer funds.

In the future, users may want to fully delegate asset management to neural networks. However, the key question remains open: who will pay compensation if the algorithm violates the established algorithm.

Cryptalist Analysis: The institutional approach to AI agents is shifting from experimentation to strict access control — banks are effectively forming a new "zero trust" model toward autonomous algorithms. For the crypto sector, this is a signal: projects offering transparent limits and damage compensation mechanisms will gain an advantage in the competition for capital. Legal uncertainty around liability for AI actions remains the main constraining factor for mass adoption.