Anthropic this month unveiled its most advanced artificial intelligence model so far, dubbed Mythos, and opened a restricted preview intended to stress its cyber-defensive capabilities. The company said the preview identified "thousands" of major vulnerabilities affecting "every major operating system and web browser."
Anthropic distributed access to the Claude Mythos preview under a controlled program called "Project Glasswing." Participants named by the company include major technology firms such as Amazon, Microsoft, Nvidia and Apple. In addition, Anthropic extended access to more than 40 other organizations that build or maintain critical software infrastructure.
Security specialists who have reviewed the announcement cautioned that Mythos' combination of advanced code generation and autonomous functionality could enable the rapid identification and exploitation of previously unknown software flaws - at a speed that may outpace the capacity of product teams to develop and deploy fixes. Industry observers highlighted the particular vulnerability of sectors that run complex, interconnected and long-lived technology stacks, citing banking as an example of an industry that could be exposed.
Anthropic itself noted that while the model can find software defects at scale in ways that may benefit defenders, the same capability, if misused, could present material risks to economies, public safety and national security.
The market reaction was immediate: U.S. software equities fell on April 9 after the Mythos preview, first disclosed on April 7, renewed worries about how AI advances could disrupt traditional technology firms.
Government and regulatory bodies have taken an active interest. The White House held discussions with Anthropic's leadership, focusing on options for collaboration, cyber resilience and how to balance ongoing AI innovation with safety considerations. Those engagements occurred even as the Pentagon applied a formal supply-chain risk designation to Anthropic.
Officials also briefed major financial institutions. Senior U.S. officials met with CEOs of large banks to outline potential risks stemming from the model and to discuss preparedness. Separately, U.S. agencies are said to be planning to obtain a version of Mythos for use by major federal bodies.
Across the Atlantic, British authorities convened talks with major banks and cybersecurity officials to evaluate possible exposures. Banking leaders in Europe are communicating closely with their regulators in light of the model's capabilities; Christian Sewing, president of the German banking association and CEO of Deutsche Bank, confirmed that banks are in active contact with European regulators about the issue.
As access to Mythos remains limited to a controlled set of industry and infrastructure partners, stakeholders - from software vendors to regulators and large financial institutions - are working to understand how to harness its defensive potential while managing the risks of misuse. The speed at which the model can surface vulnerabilities has reintroduced questions about the interplay between AI-driven discovery and the operational ability of organizations to remediate widespread software flaws.