Claude Mythos Preview has burst onto the tech scene, generating global concern despite being available for only a few weeks. Its developer, Anthropic, maintains strict restrictions on the model: only a select group of companies has been able to test it under real conditions. The official purpose is for these companies to identify vulnerabilities before they are exploited by third parties, but the exclusivity has raised alarms: if Mythos is so effective at detecting security flaws, the threat is potentially universal. Among the most concerned are major financial institutions, including the European Central Bank (ECB). Restricted access, under the name Project Glasswing, has been granted only to large U.S. corporations like Amazon, Apple, Microsoft, Alphabet, and financial entities like JP Morgan. This decision has turned Mythos into a geopolitical asset, leaving European institutions outside the circle of trust. What distinguishes Mythos is its ability to analyze code and discover "zero-day" vulnerabilities, flaws unknown even to the developers themselves, which can affect critical infrastructures like banks or energy companies. Traditionally, detecting these errors required highly qualified expert teams, but Mythos can identify them and generate code to exploit them in seconds. In response to this scenario, the ECB has called on risk managers from major Eurozone banks, such as Santander, BBVA, CaixaBank, and Sabadell, to detail their contingency plans in the face of Mythos's potential emergence. The focus is no longer just on economic risks, but on how to respond if the model falls into the hands of cybercriminals capable of causing massive data and fund thefts. The U.S. exclusivity has strained international relations. While the White House and the U.S. Treasury hold meetings with their banks, some Russian media have described Mythos as a threat "worse than a nuclear bomb." Anthropic's unilateral control over access to this cybersecurity tool poses considerable risks, especially for countries and organizations with less robust systems. The UK, through the AI Security Institute, has already accessed Mythos and confirmed its ability to execute unprecedented attacks on other AIs. Anthropic plans to expand access to British financial institutions, while the European Union continues to wait. On the other hand, Anthropic has acknowledged that unauthorized users may have accessed a version of Mythos, increasing concerns about potential leaks. Experts warn that it is only a matter of time before other powers, like China, develop similar models. OpenAI, for example, already has GPT-5-5 Cyber, an AI with advanced cybersecurity capabilities, also under restricted access.