Anthropic’s ‘Highly Capable’ AI Model Claude Mythos Emerges, Recognized as Biggest Cybersecurity Threat

Claude creator Anthropic is developing a new AI model called Claude Mythos, described internally as the industry’s most powerful model to date, with documentation of the system being released online this week.

The presence of the model was the first report in Chance Thursday after unpublished files tied to the Anthropic blog were found in public domain. Anthropic spokesperson confirmed the model’s existence in print.

“We’re building a brand that’s always moving forward in terms of design, documentation, and cyber security,” an Anthropic spokesperson said. Chance. “Based on the strength of its capabilities, we are being deliberate in the way we release it. As is the case with all companies, we are working with a small group of early access customers to test the prototype. We see this prototype as a step change and the technology we have developed to date.”

In a reserved development page reviewed by DecryptAnthropic called Mythos “the most powerful AI model we’ve ever built.”

“Mythos is a new name for a new model: bigger and smarter than our Opus models – which were, until now, the most powerful,” wrote Anthropic. “We chose this name to evoke the deep connections that connect knowledge and emotion.”

According to Anthropic, the Mythos scored “significantly higher” than the Claude Opus 4.6 in tests of software applications, academic reasoning, and cyber security.

The Mythos leak appears to have come from data stored in an unsecured system. According to ChanceAnthropic blocked access to the data store after being notified that the files were being searched online. The company said it was caused by human error in the maintenance of its CMS tools.

However, Anthropic notes named Mythos as one of the new models, and described the second model internally as “Capybara,” which the company also placed on top of their top Opus models.

The document also highlighted potential cybersecurity concerns.

“While Mythos is currently more advanced than any other AI model in cyber technology, it points to emerging models that can exploit threats in ways that exceed those of defenders,” the company wrote.

Due to these threats, the company said it plans to release the model carefully, starting with a limited release aimed at organizations working on cybersecurity.

Anthropic did not immediately respond Decrypt’s ask for feedback.

When Anthropic took down the blog post, news of the leak quickly spread through financial markets.

Shares of several cybersecurity companies he has fallen after reports surfaced, including Palo Alto Networks (PANW), which fell nearly 7%, and CrowdStrike (CRUD), which was down about 6.4%. Meanwhile, Zscaler (ZS) fell by about 5.8%, and Fortinet (FTNT) fell nearly 4% in Friday trading, according to Yahoo Finance.

The Selloff reaction is similar to the market’s reaction to the unveiling of a new Anthropic product. In February, Anthropic uncovered Claude Cowork, an AI system designed to automate work-related problems, including contract evaluation and compliance – led to significant sales in software and professional staffing companies.

The sale wiped about $285 billion off the market value as investors reassess the long-term impact of AI assistants on enterprise software businesses.

“The market response was a sign, not that AI agents will replace these businesses, but that investors have priced in the risk that providers of basic models can now compete directly with the software,” Nexatech Ventures founder Scott Dylan said. Decrypt at that time. “It’s a polite way of saying that if Anthropic can build operations management tools in-house, what’s stopping them from doing the same in finance, procurement, or HR?”