Anthropic's Leaked Claude Mythos Is a New Model Tier Above Opus — and They Say It Makes Cyberattacks Much More Likely
A configuration error in Anthropic's content management system exposed ~3,000 unpublished assets, including a draft blog post revealing Claude Mythos — a new model tier the company calls 'a step change in capabilities' and its 'most capable model to date.' Mythos introduces 'Capybara' as a new tier name sitting above Opus. Early access customers are already testing it. In parallel, Anthropic is privately briefing U.S. government officials, warning that Mythos makes large-scale cyberattacks significantly more likely.
Original sourceIn late March 2026, a misconfigured Anthropic CMS exposed nearly 3,000 unpublished pages — including a draft announcement for Claude Mythos, a model tier the company had not yet publicly acknowledged.
The draft described Capybara as "a new name for a new tier of model: larger and more intelligent than our Opus models — which were, until now, our most powerful." Anthropic confirmed the existence of the model after the leak, stating it represents "a step change" in AI performance and is currently in early access testing with select customers.
The cybersecurity angle is unusual. According to reporting, Anthropic's internal documents describe Mythos as "farther ahead of any other AI model in cyber capabilities" and the company is privately briefing top U.S. officials, warning that the model makes large-scale cyberattacks more likely in 2026 and beyond. The draft blog post itself reportedly raised unprecedented cybersecurity concerns.
No public API, no pricing, and no confirmed release date have been announced. Anthropic has not publicly addressed the data exposure beyond confirming the model's existence. The leak was caused by a configuration error in their content management system — an ironic stumble for a company staking its identity on careful, safety-conscious AI development.
The Mythos name follows Anthropic's pattern of mythological codenames. Whether 'Capybara' sticks as a product tier — or becomes a curiosity in the company's internal naming history — depends on how and when the model ships publicly.
Panel Takes
“Early access is already live for select customers, which means API access is coming. The coding and reasoning improvements Anthropic highlighted suggest Mythos will meaningfully change what's possible in agentic coding pipelines. Get on the waitlist now.”
“An accidental data leak is not a product announcement. Anthropic privately briefing the government about their own model's danger without a public safety disclosure is exactly backwards from responsible AI development. The 'step change' framing may be marketing calibrated to the leak, not the actual capability delta.”
“A new model tier sitting above Opus — with cybersecurity implications serious enough to brief U.S. officials — marks a genuine threshold moment. If Mythos delivers, it reshapes the frontier model hierarchy and proves that capability and danger are scaling together faster than governance can respond.”