Meet the Claude Mythos: A leaked Anthropic post reveals a powerful upcoming model
An accidental leakage there is now officially confirmed by AI company Anthropic about its most powerful AI model yet.
The model, now known as “Claude Mythos,” was first revealed in a report from Good luck. Anthropic has confirmed details about the store leak.
The leak includes information about the upcoming release of the Claude Mythos AI model, a “special CEO event,” and other internal assets such as PDFs and images, according to Good luck. In total, the leak includes nearly 3,000 assets that Anthropic had not previously published.
Claude Mythos leak details
According to Anthropic, the leak was the result of a problem with the content management system (CMS) used by the company. The data is uploaded by the company to the CMS. However, Anthropic failed to mark the items as private and so the data was stored in a publicly accessible data pool.
Anthropic is looking to hire a weapons specialist. It’s not what you think.
The other data was unused assets from previous announcements from Anthropic. Some of the leaked data was internal company information such as employee details or information about the CEO’s invitation-only event.
Also found within the leaked data, however, is information about Claude Mythos in an unpublished draft of a blog post.
Mashable Light Speed
Claude Mythos and Capybara
According to a leaked document about Claude Mythos, Anthropic has completed training the model and refers to it as “the most powerful AI model we have ever developed.”
In a statement provided to Fortune after the leak, an Anthropic spokesperson described Claude Mythos as a “step change” for AI and said the model is “the most capable we’ve built to date.” Anthropic says that Claude Mythos is currently in the testing phase and is available to select “early access customers.”
Anthropic sues the Pentagon as Claude’s downloads increase
In addition to the new AI model, the leaked post also mentions a new, unreleased AI model class from Anthropic called Capybara.
Anthropic currently offers three different tiers for each of its AI models: Opus, Sonnet, and Haiku. Opus is currently its strongest and most expensive category while Haiku is the smallest and least expensive. Capybara will live even beyond Opus as part of Anthropic’s largest and most intelligent AI model.
Cyybara cybersecurity risks
The leaked post also explains Anthropic’s concerns about the cybersecurity risks that Claude Mythos brings to the table, too. It seems that Anthropic believes that Claude Mythos can be used by hackers to carry out cyberattacks.
“In preparing to release Claude Capybara, we want to act with extra caution and understand the risks involved – even beyond what we learn from our own testing,” reads the leaked Anthropic post. “In particular, we want to understand the potential risks in the near term of the model in the area of ​​cybersecurity – and share the results to help cyber defenders prepare.”
Anthropic described its unreleased model as “currently ahead of any other AI model in cyber capabilities” and that the model “shows the next wave of models that can exploit vulnerabilities in ways that far exceed defenders’ efforts.”
The company said in a leaked draft blog that it is providing early access to organizations to give them “a head start on improving the resilience of their code against the next wave of AI-driven exploits.”
