Public Anthropic draft posts described Claude Mythos as the company's most powerful model and placed a new Capybara tier above Opus 4.6. The documents also point to cybersecurity capability and compute cost as rollout constraints.
The first hard signal was operational, not promotional. According to the leak thread, Anthropic’s CMS exposed “publicly accessible draft blog posts and assets,” including a Mythos page and other unpublished materials. Fortune then reported that Anthropic had left “details of upcoming product announcements” in an unsecured data store and that the company acknowledged it is “developing and testing” a new model with early-access customers Fortune confirmation.
That matters because it separates two claims engineers should treat differently. The existence of exposed draft materials appears well supported by the screenshots and CMS leak reports leak thread, while the stronger product claims should be read as draft-language plus partial company confirmation through Fortune rather than a formal launch post. The deleted post screenshot and follow-up report both reinforce that the draft page was taken down after discovery.
The leaked assets describe Mythos as a model beyond Anthropic’s current top tier. One screenshot says “Mythos is a new name for a new tier of model: larger and more intelligent than our Opus models,” and another asset shows a selector with “v1 Mythos” and “v2 Capybara,” which is why outside observers are using both names even though the exact branding is still muddled tier screenshot naming confusion.
The most concrete capability claim is the comparison target: Opus 4.6. In both the leaked screenshots and Fortune-backed summaries, Mythos is said to post “dramatically higher scores” on software coding, academic reasoning, and cybersecurity than Anthropic’s previous best model Fortune excerpt capability summary. For engineers, that suggests Anthropic is positioning Mythos less as a routine Opus refresh and more as a separate premium class with materially different cost-performance tradeoffs.
The leaked rollout copy is unusually explicit about operational limits. Anthropic says it wants “extra caution” around Mythos’s cybersecurity risk, calls the model “very expensive” to serve and use, and says it is “working to make the model much more efficient before any general release” rollout excerpt. That is stronger deployment language than a normal gated preview.
The same draft says release will start with “a small number of early-access customers” focused on cybersecurity applications, so the first real-world use case may be defensive testing rather than general developer access rollout excerpt. Community screenshots also point to internal concern about advanced cyber capability, including language about near-term cyber risk and exploit discovery, though those sharper claims come from leaked excerpts rather than an official safety report risk claims cyber defense rollout.
Taken together, the leak points to a frontier-model release plan shaped by three constraints at once: benchmark gains in coding and reasoning, serving economics, and a belief that cyber capability could outrun ordinary release controls Fortune confirmation.
Anthropic’s website content management system included publicly accessible draft blog posts and assets, including Claude Mythos and Anthropic CEO Forum
Claude Mythos Blog Post Saved before it was taken down. m1astra-mythos.pages.dev
Confirmed by Fortune: Opus 5: so good that it poses a danger. It's a massive leap from version 4.6, and it's so advanced that it poses a huge security risk. They'll roll it out slowly because they're concerned about unforeseen consequences, especially security risks. Besides Show more
Holy sh*t huge leak Claude Opus 5 "Mythos" leak: Claude Mythos, a new tier beyond Opus models, delivering major performance gains compared to Opus 4.6 in coding, academic reasoning, and cybersecurity, and signaling a step-change in frontier AI capability. The model is so
