The key facts
- Citibank, JPMorgan, Bank of America and PNC have updated their terms of service, placing full responsibility for the actions of third-party AI agents on clients.
- Banks refuse to compensate for losses from hacks or errors if they occurred through tools to which the client granted access.
- The decision reflects banks' desire to retain control over financial flows and protect their margins from competition with AI startups.
US Banks Shift Liability for AI Agent Actions
The largest commercial banks in the US, including Citibank, JPMorgan, Bank of America and PNC, have updated their user agreements, stating that clients who use third-party artificial intelligence tools to manage their finances assume full responsibility for any losses. This applies both to unintentional transactions performed by automated assistants and to cases of account hacking or credential leaks through these tools. Citibank's updated documents explicitly state that clients who have granted access to their accounts to third-party AI agents are responsible for all actions performed through them, even if access was obtained by attackers due to a vulnerability in the AI service itself.
Conflict of Interest and Control over Financial Flows
The banks' decision is driven not only by cybersecurity concerns but also by the desire to retain control over the customer interaction channel, which AI startups see as a new monetization opportunity. Financial institutions fear that automated agents will help consumers avoid fees and find better terms, which would reduce banks' margins. In response, Citibank notes that clients may use AI agents provided by the bank itself without the risk of incurring additional liability. Against this backdrop, some developers, such as Anthropic, already position their financial tools exclusively in a 'read-only' mode, excluding the possibility of making transactions or managing assets.

