US Central Bank Warns of Security Risks from Anthropic's AI Model – Access to Mythos AI Unavailable for Months

As CNBC Top News reports (https://www.cnbc.com/2026/07/21/fed-mythos-ai-cybersecurity-banks-project-glasswing.html), the US Federal Reserve (Fed) has warned of potential security risks posed by Anthropic's AI model Mythos. At the same time, the Fed was unable to access the system itself for months until mid-July 2026 to conduct its own analyses and security assessments.

Background on Mythos AI and Security Concerns

Anthropic, a leading company in artificial intelligence, developed Mythos AI as an advanced language model platform. The Fed expressed concerns that such AI systems, increasingly used in financial institutions and critical infrastructures, could present potential vulnerabilities to cyberattacks. Especially in the banking sector, which processes sensitive customer data and financial transactions, the security of AI applications is of central importance.

Fed Without Access Despite Growing Risks

According to CNBC, however, the Fed had no access to the preview version of Mythos AI until mid-July 2026, which complicated its own security evaluations. While other financial institutions and technology companies had already taken measures to harden their systems against possible AI-related attacks, the Fed remained cautious in this regard. This raises questions about coordination between regulatory authorities and technology providers.

Project Glasswing – Fed Relies on Own Cyber Defense

Parallel to concerns about external AI systems, the Fed is working on "Project Glasswing," an internal cybersecurity program aimed at increasing the resilience of financial infrastructure against modern threats. This project is intended to reduce dependence on external AI providers and strengthen control over security-critical systems.

Why It Matters

The increasing integration of AI technologies into the financial sector brings significant opportunities but also new risks. The Fed, as the central regulatory and supervisory authority, plays a key role in defining and enforcing security standards. That the Fed itself lacked access to an important AI model for months underscores the challenges in regulating and monitoring rapidly evolving technologies. For banks and financial service providers, this means they must evaluate their AI applications not only for efficiency and innovation but also for security and compliance. The Fed situation highlights the importance of transparency and collaboration among developers, users, and regulators to identify and minimize risks early.

Outlook

The Fed is expected to further expand its capabilities to assess and monitor AI systems. At the same time, the industry is likely to increase investments in security solutions to meet supervisory requirements. Developments around Mythos AI and Project Glasswing could serve as a blueprint for managing AI risks in the financial sector. Overall, the case illustrates that technological innovations like AI offer enormous potential but also introduce a new dimension of cyber risks that must be carefully managed.