Anthropic’s ‘Most Capable’ AI Model Claude Mythos Leaks, Deemed Major Cybersecurity Threat

Claude maker Anthropic is developing a new AI model called Claude Mythos, described internally as the company’s most capable model yet, with draft material about the system leaked online this week.

The existence of the model was the first reported Through Fortune Thursday after unpublished files associated with Anthropic’s blog were discovered in a publicly accessible data cache. An Anthropic spokesperson confirmed the model’s existence to the publication.

“We are developing a general-purpose model with meaningful advances in reasoning, coding and cybersecurity,” an Anthropic spokesperson told Fortune. “Given the power of its capabilities, we are being deliberate about how we release it. As is industry standard, we are working with a small group of early access customers to test the model. We view this model as a step change and the most capable we have built to date.”

In an archived development page rated by DeclutterAnthropic called Mythos “the most powerful AI model we have ever developed.”

“Mythos is a new name for a new model level: bigger and more intelligent than our Opus models – which were our most powerful to date,” wrote Anthropic. “We chose the name to evoke the deep connective tissues that connect knowledge and ideas.”

According to Anthropic, Mythos scored “dramatically higher” than Claude Opus 4.6 on tests of software coding, academic reasoning, and cybersecurity.

The Mythos leak appears to come from draft material stored in an unsecured content management system. According to FortuneAnthropic restricted public access to the data store after being notified that the files were searchable online. The company attributed exposure to human error in the configuration of its CMS tools.

However, Anthropic’s documents labeled Mythos as version one of the new model, and version two was described internally as ‘Capybara’, which also placed the company above its current top-of-the-line Opus models.

The draft material also raised concerns about the system’s potential impact on cybersecurity.

“While Mythos is currently far ahead of any other AI model in cyber capabilities, it predicts a coming wave of models that can exploit vulnerabilities in ways far beyond the efforts of defenders,” the company wrote.

Because of these risks, the company said it plans to release the model cautiously, starting with a limited early access rollout aimed at organizations working on cybersecurity defenses.

Anthropic did not immediately respond Decode request for comment.

While Anthropic removed the blog post, news of the leak quickly spread to the financial markets.

Shares of various cybersecurity companies decreased after the reports surfaced, including Palo Alto Networks (PANW), which fell about 7%, and CrowdStrike (CRWD), which fell by about 6.4%. Meanwhile, Zscaler (ZS) fell about 5.8%, and Fortinet (FTNT) fell about 4% in Friday trading, according to Yahoo Finance.

The sell-off reaction reflects a similar market reaction to the unveiling of a new Anthropic product. In February, Anthropic unveiled Claude Cowork, an AI system designed to automate complex workplace tasks – including contract review and compliance – which led to a broad sell-out among software and professional services firms.

This sell-off wiped out approximately $285 billion in market value as investors reevaluated the long-term impact of AI agents on enterprise software companies.

“The market reaction was a signal not that AI agents will immediately replace these companies, but that investors are finally pricing in the structural risk that foundation model providers can now compete directly with the software layer,” Scott Dylan, founder of Nexatech Ventures, told me. Declutter at the time. “That’s a polite way of saying that if Anthropic can build a legal workflow tool internally, what’s stopping them from doing the same in finance, procurement or HR?”