Major US banks are developing response plans for severe AI incidents, focusing not only on preventing attacks but also on maintaining operations, protecting accounts, and restoring customer trust after a crisis.
According to Axios, many banks have detected unverified AI agent access activities, primarily driven by customers providing their login credentials. In some cases, AI agents appeared to perform actions beyond the users’ requests.
Risks from AI agent account access differ from the threat of cyberattacks utilizing advanced AI models. Banks are building defense strategies for both risk groups.
Recent findings regarding severe security vulnerabilities powered by AI have heightened concerns over cybersecurity crises. Salesforce CEO Marc Benioff has issued an urgent warning signal.
US politicians are calling for increased regulation of AI agents. Former White House AI advisor David Sacks proposed a mechanism allowing users to grant, limit, and revoke access rights, initially restricting AI to read-only data access.
JPMorgan and other banks have implemented measures to block AI agent access while exploring secure connections through open banking platforms.
JPMorgan CEO Jamie Dimon warned that AI agents could execute transactions contrary to customers’ wishes. An economist even hypothesized that AI agents could simultaneously transfer funds to higher-yielding accounts, creating the risk of a mass bank run.
However, banks affirm they have found no evidence of AI agents spinning out of control or launching attacks that cause damage to customer accounts and funds.
📌 Conclusion: Wall Street is shifting from AI risk mitigation to preparing for crisis response and recovery. The focus is on controlling AI agent access, protecting customer assets, and maintaining banking operations. JPMorgan advises strictly against sharing passwords with AI agents, prioritizing secure open banking connections. The governance challenge is distinguishing user-authorized activities from cyberattack behavior, while establishing appropriate access granting and revocation mechanisms.
