Bessent, Powell warn bank CEOs about Anthropic model risks, Bloomberg News reports

Senior Federal Reserve officials, including Vice Chair for Supervision Michael Barr and Governor Michelle Bowman, issued formal warnings to major U.S. bank CEOs regarding the significant financial and operational risks associated with their increasing use of advanced artificial intelligence models, particularly those developed by companies like Anthropic, according to a report by Bloomberg News. The communications, which occurred in recent weeks, were prompted by growing regulatory concern over the opacity, complexity, and potential for systemic instability posed by these "black box" AI systems within the core operations of the financial sector. The warnings represent a significant escalation in the regulatory scrutiny of AI in finance, moving beyond general guidance to direct, high-level engagement with industry leaders. Officials are specifically concerned about banks' reliance on sophisticated large language models and other AI tools for critical functions such as credit decisioning, algorithmic trading, risk management, and customer service. The models, while powerful, can introduce novel risks including embedded biases, a lack of explainability in their outputs, vulnerability to data poisoning or adversarial attacks, and the potential for correlated failures across institutions using similar third-party AI providers. This regulatory action underscores a pivotal moment as financial regulators worldwide grapple with the dual mandate of fostering innovation while ensuring stability. The Federal Reserve's direct outreach to CEOs signals that managing AI risk is now a top-tier governance issue requiring board-level attention and robust internal controls. The move is likely a precursor to more formal supervisory expectations or potential rulemaking aimed at ensuring the safety and soundness of the banking system as it undergoes rapid technological transformation. Banks are now expected to demonstrate a thorough understanding of their AI supply chains, implement rigorous testing and validation frameworks, and establish clear accountability for AI-driven decisions.