Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks

Anthropic, the artificial intelligence safety company, has deployed its new Mythos AI model to a limited group of major U.S. financial institutions, including JPMorgan Chase and Citigroup, following discussions between Anthropic executives and bank security chiefs in late 2024. The controlled rollout was initiated due to significant concerns that sophisticated cybercriminals or state-sponsored hackers could weaponize advanced AI capabilities to orchestrate complex financial attacks, manipulate markets, or breach critical banking infrastructure. The decision to engage directly with top-tier banks came after high-level briefings where Anthropic's leadership, including figures like Dario Amodei, presented the potential dual-use nature of the Mythos model. While designed for defensive cybersecurity applications—such as simulating attacks to find network vulnerabilities or analyzing malware—the model's advanced reasoning and code-generation abilities also represent a potent offensive tool if obtained by malicious actors. This preemptive, confidential sharing with trusted financial partners represents a novel approach in the AI industry, prioritizing security over broad commercial release. This move underscores the escalating arms race in AI-powered cybersecurity within the financial sector, which remains a prime target for digital crime. By providing early access, Anthropic aims to help these institutions harden their defenses against the very threats its technology could theoretically enable, creating a feedback loop to improve the model's safety features. The action also reflects growing regulatory and ethical pressures on AI developers to implement stringent guardrails and controlled deployment strategies for their most powerful systems, especially those with clear national security and economic stability implications.