Concerns are mounting over the cybersecurity risks posed by Anthropic’s AI tool, Claude Mythos Preview, as it rapidly identifies software vulnerabilities in critical systems worldwide. Sam Woods, chief executive of the Prudential Regulation Authority, highlighted these cyber risks as his top concern for the banking sector during a recent interview.

Claude Mythos, a large language model designed to detect software weaknesses, has become a focal point in cybersecurity discussions since its launch. While organisations and governments have been leveraging the tool to expose and address vulnerabilities within their software, experts warn that the primary threat may stem from Anthropic itself being targeted by cyberattacks.

As of late May, Mythos had flagged over 23,000 potential vulnerabilities in open-source projects alone. Out of these, approximately 1,600 were reported to maintainers, but fewer than 100 had been patched, underscoring the scale of unresolved security issues. Critical infrastructure providers, major financial institutions, and software vendors trusted by governments and enterprises have increasingly submitted their codebases to Mythos for analysis. This extensive usage has positioned Anthropic as a central repository for sensitive intelligence on zero-day vulnerabilities affecting essential systems worldwide.

The concentration of such data at Anthropic has raised alarm among cybersecurity professionals and policymakers. Past incidents involving the exposure of sensitive vulnerability information have led to widespread attacks, such as the WannaCry ransomware outbreak linked to leaked NSA cyber tools distributed by the hacker group known as the “Shadow Brokers.” These precedents illustrate the potential widespread damage that could result if Anthropic were compromised.

In light of these concerns, calls are growing for governments and international organisations to scrutinise who can access the vulnerability data stored by Anthropic, evaluate the safeguards protecting this information, and contemplate the possible consequences of a significant breach. With Anthropic becoming a critical security dependency for some of the world’s most important systems, questions about data protection and risk mitigation are increasingly urgent.