Exclusive: Anthropic 'Mythos' AI model representing 'step change' in capabilities revealed in data leak

Anthropic is developing a new AI model more capable than any previously released. The company said it has begun testing the model with early access customers following a data leak that revealed its existence. Descriptions of the model were stored in a publicly accessible data cache and reviewed by Fortune. An Anthropic spokesperson said the new model represents a step change in AI performance and is the most capable the company has built to date. The model is referred to as Claude Mythos in a draft blog post found in the cache. Nearly 3,000 assets linked to the blog were publicly accessible in the data store. The leak resulted from human error in the configuration of the company's content management system. Anthropic acknowledged the error and removed the public's ability to search the data store after being notified. The draft post also discussed the model under the name Capybara as a new tier larger than Opus models. The document said the company had completed training of Claude Mythos, described as the most powerful AI model developed to date. It noted dramatically higher scores on tests of software coding, academic reasoning and cybersecurity compared to previous models.