Fortune broke the story on March 26, 2026. Security researchers Roy Paz of LayerX Security and Alexandre Pauwels of the University of Cambridge independently discovered an unsecured data cache containing approximately 3,000 unpublished Anthropic assets, including two versions of the same blog post that differed in only one detail: the model’s name. One draft called it Claude Mythos. The other called it Capybara.
The company confirmed the leak was real. A spokesperson told Fortune the company is “developing a general purpose model with meaningful advances in reasoning, coding, and cybersecurity” and described it as “a step change” in AI performance, “the most capable we’ve built to date.” The company attributed the exposure to human error in its content management system configuration.
What the leaked drafts describe is a new model tier above Opus. According to The Decoder’s analysis of the documents, the model achieves “dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others” compared to Claude Opus 4.6. The documents position it as “larger and more intelligent than our Opus models, which were, until now, our most powerful.”
The cybersecurity dimension is where this gets serious. Anthropic’s own internal assessment, now public, states the model is “currently far ahead of any other AI model in cyber capabilities” and warns it “presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.” That’s not an external critic raising alarms. That’s Anthropic’s own language about its own model.
The rollout plan visible in the leaked documents reflects that concern. The rollout starts with a small group of early-access customers specifically tasked with evaluating cybersecurity applications, with broader API access expanding gradually. The company also acknowledged the model is “very expensive for us to serve, and will be very expensive for our customers to use,” and said it’s working to make it “much more efficient before any general release.”
The naming question, Mythos versus Capybara, appears to be genuinely unresolved. A spokesperson told Fortune the documents were “early drafts of content that were being considered for publication,” which is consistent with a company that hasn’t finalized branding on a product it hasn’t officially announced.
For practitioners, three things matter here. First, a model tier above Opus changes the competitive landscape. Google’s Gemini Ultra and OpenAI’s rumored next-generation models now have a benchmark to clear. Second, the cybersecurity warning is unusually direct for a company describing its own product, and it suggests Anthropic’s safety team has significant influence over release strategy. Third, the leak itself, nearly 3,000 internal documents exposed through a CMS misconfiguration, is an operational security failure that undercuts the careful-release narrative.
No release date has been announced. What we know is that the model exists, it’s in testing, and Anthropic considers it dangerous enough to warrant a staged rollout starting with security researchers. Watch for an official announcement in the coming weeks.