Anthropic's Mythos Leaked — And the Real Story Isn't the Model

On March 26, Fortune broke a story that made the rounds fast: Anthropic has been training a new model called Claude Mythos (also referred to internally as "Capybara"), and it leaked through a misconfigured content management system. Not through a sophisticated attack. Not through an insider. Through a publicly searchable data cache that contained ~3,000 unpublished blog assets. Let that sink in for a second. What We Know About Mythos Anthropic confirmed they're testing a new model with "early access customers" and called it "a step change" and "the most capable we've built to date." The leaked draft blog post describes: A new tier called Capybara , sitting above Opus (their current largest tier) "Dramatically higher scores" on coding, academic reasoning, and cybersecurity benchmarks vs Claude Opus 4.6 An acknowledgment that the model poses "unprecedented cybersecurity risks" That last point is interesting. Anthropic has been increasingly vocal about their Responsible Scaling Policy, an