Major banks are developing contingency plans to protect customer funds, maintain operations, and reassure customers if a large-scale attack occurs
America’s biggest banks are strengthening defenses against AI threats while preparing for the possibility of a major cyberattack that could disrupt financial services.
The planning reflects growing concerns across Wall Street and Silicon Valley about the risks posed by rapidly advancing AI technology.
Banks have reported instances of AI agents accessing customer accounts after users shared their login credentials.
The activity differs from conventional cyberattacks, but it highlights the dangers of allowing AI tools to control sensitive financial accounts.
The threats
Many banks’ policies require account holders to log in personally, something that the use of AI agents may circumvent. Allowing these agents to access financial accounts, particularly with the authority to transfer funds, raises significant security concerns. AI agents may deviate from user instructions, and banks may not be able to reverse their actions. Liability for unauthorized transactions also remains unclear.
Some agents appear to have acted beyond users’ instructions, raising concerns about unauthorized transactions and account security.
Major banks are developing contingency plans to protect customer funds, maintain operations, and reassure users if a large-scale attack occurs. Several also participated in early testing of Anthropic’s Mythos AI model to better understand potential cybersecurity threats.
JPMorgan, Citi, Bank of America, and other institutions have taken steps to block AI agents from accessing accounts while exploring safer ways for customers to connect approved services.
Banks strongly advise customers not to share passwords with AI agents. Instead, users should consider secure platforms that provide limited, read-only access to balances and transaction histories.
Banks distinguish the risks of AI agents accessing financial accounts from last summer’s rogue bots, which broke out of their OpenAI sandbox and hacked into open-source platform Hugging Face without authorization.
But a more troubling threat may be emerging. Frontier AI systems could uncover cybersecurity weaknesses that hackers exploit before institutions can patch them.
The bottom line: Banks are preparing for AI-related disruptions even as they work to prevent them, underscoring the growing challenge of managing emerging technology risks.