Bessent called in US bank CEOs to discuss Anthropic model’s cyber risks

US Treasury Secretary Janet Yellen convened an urgent meeting with the CEOs of major American banks in Washington, D.C., on Tuesday, to discuss significant cybersecurity risks identified by Anthropic's latest AI model. The meeting was prompted by the AI system's discovery of previously undetected, decades-old vulnerabilities within critical financial infrastructure, raising immediate concerns about systemic stability. The gathering represents a proactive, high-level government response to a novel threat vector emerging from advanced artificial intelligence. The core of the discussion centered on findings from Anthropic's Claude model, which reportedly uncovered legacy software flaws and architectural weaknesses in banking systems that have persisted for years without detection by conventional security audits. These vulnerabilities, some embedded in foundational code from the 1990s and early 2000s, could potentially be exploited to disrupt payment networks, compromise customer data, or manipulate financial markets. Secretary Yellen emphasized the need for a coordinated, sector-wide assessment and remediation plan, framing the issue as both a national security and economic priority. This incident marks a pivotal moment in the regulatory understanding of AI's dual-use nature in the financial sector. While banks have aggressively adopted AI for fraud detection and customer service, the Anthropic model's capability to autonomously find critical flaws highlights a new risk: AI systems can be used to probe and expose systemic weaknesses just as effectively as they can be used to defend them. The Treasury Department is now reportedly evaluating new guidelines for AI security testing and vulnerability disclosure in critical infrastructure, signaling a shift toward more anticipatory and AI-aware financial regulation. The meeting's outcome includes a directive for banks to conduct immediate, AI-assisted penetration tests on their core systems and to establish a confidential information-sharing channel with Treasury and cybersecurity agencies. The episode underscores the accelerating convergence of financial stability policy and technological governance, as regulators scramble to keep pace with AI capabilities that are evolving faster than traditional risk frameworks.