Anthropic Plans to Grant European Banks Access to Mythos AI Model

Anthropic is preparing to open access to its high-powered and controversial AI model, Mythos, to European banks within the coming days or weeks, Reuters reports, citing informed sources. Initial access to the model had previously been restricted to a select group of major U.S. financial institutions.

The Mythos model has become one of the most discussed topics among cybersecurity inner circles. Designed to handle highly specialized and mission-critical tasks, Mythos represents a next-generation model engineered for complex logical reasoning and digital infrastructure analysis. Its core competency lies in the instantaneous detection of vulnerabilities in software code. Where a team of programmers might require weeks of manual labor, Mythos can complete the task in mere minutes.

Rather than simply scanning data, the model performs a deep audit, methodically identifying hidden security "holes" and critical weak points, even within the most complex and sprawling architectures of large-scale networks.

Of particular value is its proactive capability: the model effectively recognizes "zero-day" threats, pinpointing potential entry points that remain unknown to the global cybersecurity community. Mythos does not stop at observation; shifting into an "active defense" mode, the system immediately generates detailed recommendations and autonomously writes software patches, effectively sealing security breaches in real-time.

However, the power of Mythos is a double-edged sword. The model is capable of autonomously generating working exploits, simulating sophisticated attacks—such as a 32-step breach of a corporate network—and accelerating the development of hacks. This has alarmed regulators, who view the tool as a systemic risk for banks operating with legacy software. Institutions such as Barclays and Deutsche Bank have characterized it as a "serious threat," noting that bad actors equipped with similar unrestricted models could strike faster than defenders can respond.

Anthropic acknowledges this duality, stating the model is "too powerful and dangerous" for an open public release. Consequently, access is granted only to vetted banks for the purpose of proactive defense. Nevertheless, experts warn that the window for patching is shrinking as AI-driven attacks move toward the mainstream. Regulators are now calling for urgent stress tests to ensure defensive measures outpace the evolving threat.

Until now, access to Mythos has been tightly controlled under Anthropic’s "Project Glasswing" initiative, limited to approximately 40 organizations involved in building or maintaining critical software infrastructure.

JPMorgan Chase remains the only bank Anthropic has publicly confirmed as a Mythos partner, though Reuters sources indicate that Bank of America has been involved in Project Glasswing since its inception. Last week, Morgan Stanley CEO Ted Pick and Goldman Sachs CEO David Solomon both confirmed access to Mythos during calls with analysts. Citigroup is also reportedly utilizing the model for internal testing.

In Europe, no financial institution has been granted access yet, according to several high-ranking sources in banking and regulatory circles. Deutsche Bank CEO Christian Sewing stated on Monday that banks are in close contact with European regulators regarding Mythos. "It is definitely something we need to consider in our day-to-day risk management—and that is exactly what we are doing," Sewing said

Follow NEWS.am Tech on Facebook and Twitter