Sometimes reality writes better stories than any PR team could. Anthropic — the company that’s built its entire brand on ‘responsible AI’ — accidentally leaked details about its most powerful model yet. And what came out is both fascinating and deeply concerning.
What happened
A configuration error in Anthropic’s content management system made nearly 3,000 unpublished assets publicly accessible. Security researchers from LayerX Security and the University of Cambridge discovered the exposed data store. Fortune reviewed the documents and notified Anthropic, which then restricted access.
Among the documents: a draft blog post about a new model called Claude Mythos.
What we know about Mythos
An Anthropic spokesperson confirmed to Fortune that the new model represents “a step change” in AI performance and is currently being trialed by early access customers. According to the leaked documents, Mythos is “the most capable model we’ve built to date.”
Here’s where it gets interesting — and a bit scary: the internal documents explicitly warn of “unprecedented cybersecurity risks.” Mythos is reportedly “currently far ahead of any other AI model in cyber capabilities” and could find and exploit vulnerabilities at speeds defenders simply can’t match.
Capybara: A new model tier
The leaks also point to a new model tier called Capybara — even larger and more capable than Opus, Anthropic’s current top-tier model. In benchmarks for software coding, academic reasoning, and cybersecurity, Capybara reportedly outperforms Claude Opus 4.6 by a significant margin.
My take
There’s a certain irony in Anthropic — the company that puts safety above everything — exposing its most sensitive secrets through a simple configuration error. At the same time, you have to respect that they’re writing honestly about the risks of their own models internally. Many companies wouldn’t even put those warnings on paper.
The real question is: when a model is so powerful that even its creators are sounding the alarm — how do we as a society deal with that? Anthropic will need to answer that question before Mythos goes wide.
Sources: