Anthropic ran an ad about Claude on connected TV in
March. Then came the rush to beat OpenAI to an initial public offering, and a U.S. federal judge’s decision to halt the Trump administration’s designation as being a supply-chain risk,
with a win for Anthropic in its lawsuit against the Pentagon.
Then on Thursday, a data leak revealed Anthropic's next powerful AI model, which the company acknowledged as being a "step change" and called the "most capable model to date."
The data leak
-- caused by a major security lapse in Anthropic's public-facing content-management system -- revealed that the company is working on a new model release.
Anthropic blamed “human
error” for the data leak. The way the CMS was configured made the draft blog post accessible for anyone to read, the company said in a statement to Fortune.
advertisement
advertisement
The material was left in an unsecured and publicly searchable data store. Anthropic described it as an “early drafts of content considered for publication.”
Nearly 3,000
pieces of content were made accessible. Alexandre Pauwels, Fortune cybersecurity researcher, discovered the unsecured data cache. The files included draft blog posts, images, and research
papers.
An Anthropic spokesperson has since acknowledged the new project as “Claude Mythos," and stated that the company had completed training the model.
The draft blog notes
that the model is expensive to run and not yet ready for general release.
“We’re developing a general-purpose model with meaningful advances in reasoning, coding, and
cybersecurity,” an Anthropic spokesperson told Fortune. “Given the strength of its capabilities, we’re being deliberate about how we release it.”
Anthropic is
working with a small group of early access customers to test the model, and consider it to be a “step change,” meaning the most capable the company has built.
The draft blog post
also revealed a new tier of AI models called “Capybara,” and described it as a “larger and more intelligent” than Anthropic’s Opus models.