Major US banks are increasingly blocking third-party AI agents from accessing customer accounts, raising concerns about cybersecurity, financial autonomy and the future of AI-driven banking.
According to recent reports, customers are sharing banking credentials with autonomous AI assistants to manage finances, analyse transactions and optimise investment decisions.
However, banks warn that allowing external agents unrestricted account access creates significant vulnerabilities, including credential theft, unauthorised transactions and financial fraud.
JPMorgan Chase CEO Jamie Dimon has highlighted the risks of allowing unverified AI agents into banking systems, emphasising customer protection and institutional security.
Critics argue that banks may also fear losing revenue as AI agents automatically transfer idle deposits into higher-yielding financial products, challenging traditional banking profitability.
The cybersecurity implications are substantial. Compromised autonomous agents could execute fraudulent transactions, manipulate payment instructions or expose confidential financial information without immediate human intervention.
Unlike conventional cyberattacks, agentic AI introduces machine-speed decision-making, potentially allowing attackers to exploit compromised credentials, malicious instructions and interconnected financial applications.
Banks must therefore move beyond traditional authentication towards continuous behavioural verification, zero-trust access, transaction monitoring and real-time anomaly detection.
Emerging security frameworks should combine restricted API permissions, read-only access, transaction limits, cryptographic authorisation and human approval for high-risk financial activities. Independent monitoring of agent behaviour is equally critical.
The bigger question is whether banks can distinguish legitimate AI assistants from malicious autonomous agents.The answer lies in establishing enforceable digital trust boundaries, where every agent, identity and transaction is continuously verified. Without these safeguards, autonomous AI could become a powerful new channel for financial cybercrime.
