Why the World’s Banks Are so Worried About Anthropic’s Latest AI Model

The rollout of Anthropic’s Mythos model marks a watershed moment in AI‑enabled cybersecurity. Internal testing revealed a trove of previously unknown zero‑day flaws that span operating systems, browsers, and network stacks. Because the model can automate vulnerability discovery at unprecedented speed, it threatens to outpace traditional patch cycles. Anthropic’s decision to restrict Mythos to a select defensive coalition—featuring tech giants like Microsoft, Amazon Web Services, Apple and Cisco—reflects a cautious approach to prevent misuse while leveraging the model’s capabilities to harden critical software.

For banks, the stakes are especially high. Legacy core‑banking platforms, many built on decades‑old code, lack the agility to respond to rapid exploit development. The IMF’s spring meeting underscored regulators’ concern that AI‑driven attacks could erode confidence in financial stability. In response, Anthropic has committed $100 million in usage credits and $4 million in open‑source grants to accelerate bug‑fixing efforts, while more than 40 organizations—including several U.S. banks—have been granted limited access for defensive testing. This collaborative model aims to close the security gap before threat actors can weaponize the findings.

The Mythos episode illustrates a broader shift: software complexity makes absolute bug‑free guarantees unrealistic, and AI is becoming both a sword and a shield. Initiatives like the EU’s age‑verification app and the Beneficial AI Foundation’s effort to mathematically verify Signal’s code show the growing need for rigorous, AI‑assisted security audits. As AI continues to evolve, the industry must balance innovation with robust safeguards, turning the same technology that uncovers vulnerabilities into a cornerstone of cyber‑defense.