A data leak has revealed that Anthropic has been developing a powerful new AI model, prompting concerns about potential risks and delaying a broader release.
According to a report by Fortune, the model referred to as Claude Mythos came to light after internal materials were accidentally exposed in a publicly accessible cache.
The leak offered a rare glimpse into what the company describes as its most advanced model to date.
When Fortune contacted the company, Anthropic acknowledged that the model exists. “We’re developing a general-purpose model with meaningful advances in reasoning, coding, and cybersecurity,” a spokesperson said.
The model is said to represent a significant leap beyond Anthropic’s existing Claude lineup, such as Opus 4.6.
The disclosure followed a security lapse in which internal materials, including a draft blog post, were inadvertently made publicly accessible through an unsecured data cache.
The documents were reviewed by Fortune and independent cybersecurity researchers before Anthropic restricted access.
According to the leaked draft, Claude Mythos is part of a new ‘Capybara’ tier, which surpasses the company’s existing Opus line in capability and cost.
The blog described it as “by far the most powerful AI model we’ve ever developed,” with significantly improved performance in software coding, academic reasoning, and cybersecurity tasks.
Anthropic said the exposure was the result of human error in configuring its content management system, describing the materials as “early drafts of content considered for publication.”
The leaked documents also highlighted serious concerns about the model’s potential misuse, particularly in cybersecurity.
Anthropic warned that the system is “currently far ahead of any other AI model in cyber capabilities” and “it presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders.”
Early information suggests the system is already being tested with a limited group of users, indicating it is in an advanced stage of development.
The company indicated that early access will prioritise organisations working on cyber defence to help prepare for such risks.
Anthropic has not formally announced the model, and details remain limited.
Meanwhile, the company is discussing plans to go public as early as the fourth quarter of 2026, with bankers expecting the startup to raise more than $60 billion in what could become one of the largest initial public offerings in history, according to a report by The Information.
Your reaction