Banks met on Anthropic risk
Central banks and major lenders reportedly met to discuss cyber risks raised by recent Anthropic model advances, with coverage noting the meeting and security concerns. Reporting also highlighted that Anthropic leadership has engaged with senior officials even amid other institutional frictions. (financialpost.com) (npr.org) (timesofindia.indiatimes.com)
The Bank of Canada met major lenders and financial firms on Friday, April 10, to discuss cyber risks tied to Anthropic’s latest artificial intelligence model. (bloomberg.com) That Canadian meeting followed a Washington session earlier in the week, where Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called in top United States bank chief executives for an urgent discussion about the same threat. (bloomberg.com) (reuters.com) The issue is Anthropic’s unreleased model, called Claude Mythos 2 Preview, which the company said can find and exploit software flaws better than almost all human researchers. Anthropic said the model has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser. (anthropic.com) (npr.org) Banks care because the same kind of hidden software flaw can open payment systems, customer records, trading infrastructure, and internal networks to attack. Anthropic said systems that run banking, medical, logistics, and power networks all rely on software that still contains serious bugs. (anthropic.com) Anthropic did not broadly release Mythos. Instead, it said access would be limited to launch partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, Nvidia, Palo Alto Networks, and more than 40 additional organizations that maintain critical software. (anthropic.com) (reuters.com) Anthropic paired that restricted rollout with Project Glasswing, a defensive program that commits up to $100 million in usage credits and $4 million in donations to open-source security groups. The company said the goal is to use the model to scan and secure critical software before similar capabilities spread more widely. (anthropic.com) At the same time, Anthropic has stayed in direct contact with senior officials in Washington. CNBC, cited by the Times of India, reported that Chief Executive Dario Amodei joined a call last week with Vice President JD Vance and Bessent alongside other technology leaders before Mythos was released to a limited group. (timesofindia.indiatimes.com) Those talks are happening while Anthropic is also fighting the Pentagon in court. A federal appeals court in Washington, District of Columbia, on April 8 denied Anthropic’s request to temporarily block a Defense Department designation that labeled the company a supply chain risk and cut it off from Pentagon contracts while the case continues. (cnbc.com) So the same company that says its model can uncover flaws across core software is now being briefed by, questioned by, and debated inside the governments that regulate banks and buy military technology. The immediate response, in both Ottawa and Washington, has been to get central banks, finance officials, and lenders in the same room before the model’s capabilities spread further. (bloomberg.com) (reuters.com)
