The new Anthropic AI model, Claude Mythos, triggered cybersecurity fears after a human error exposed draft documents describing the system, warning it could enable large-scale cyberattacks, marking a “step change” in AI capabilities.
Caused by a misconfigured content management system, the leak briefly made Anthropic’s models draft blog publicly searchable. It showed that Mythos AI, part of a new AI tier called Capybara, is larger and more intelligent than the company’s previous Opus models.
Fortune reported that the documents were stored in an unsecured data cache alongside nearly 3,000 other unpublished assets. The Anthropic AI model was noted as a key application.
AI Performance Leap
An Anthropic spokesperson told Fortune, that the company is “developing a general purpose model with meaningful advances in reasoning, coding, and cybersecurity. Given the strength of its capabilities, we’re being deliberate about how we release it. As is standard practice across the industry, we’re working with a small group of early access customers to test the model. We consider this model a step change and the most capable we’ve built to date.”
The draft blog 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.”
Capybara and Mythos appear to be the same underlying model. Compared with Claude Opus 4.6, the company says Capybara “gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others.”
Anthropic applied AI is central to these efforts, and the model is being trialed for the Anthropic AI model.
Anthropic is currently trialing Mythos with early access customers while working to improve efficiency. The model is extremely compute-intensive and expensive to run, and a general release is not yet planned. Anthropic AI tools are part of the early access workflow.
Cybersecurity Implications
Cybersecurity experts caution that Mythos could significantly amplify hacking capabilities. As Anthropic AI model, learn to act and reason with more autonomy, hackers could simultaneously run multiple sophisticated attacks, making defense more difficult. Employees using Anthropic agent skills may inadvertently create access points to corporate networks, while human identities become easier to breach, according to a Palo Alto Networks executive cited by Euronews Next.
This concern echoes a past incident: in November 2025, Anthropic’s AI model was used in an autonomous cyberattack after Claude was manipulated by Chinese state hackers.
While that incident involved Claude rather than Mythos, it demonstrates the risks of Anthropic AI use, showing that breaches could happen again.
Demanding Anthropic security ramp-up measures are now reportedly underway.
According to Axios, Anthropic has privately warned top government officials that Mythos increases the likelihood of large-scale cyberattacks in 2026.
“The new model poses unprecedented cybersecurity risks,” the draft blog stated, emphasizing the scale of potential threats. Cybersecurity stocks reportedly dropped following the rumors about Anthropic’s advanced model.
The exposure came alongside details of an invite-only CEO summit in Europe, part of Anthropic’s efforts to market its AI to large corporate clients.
LayerX Security and the University of Cambridge researchers confirmed that draft materials, including the Mythos AI blog, had been publicly accessible prior to Anthropic securing the data store after being informed by Fortune.
With Mythos, Anthropic Agentic AI demonstrates advanced reasoning and control capabilities. The model also highlights Anthropic applied AI and reinforces the company’s reputation models.
Overall, the Anthropic AI model, powered by Mythos, signals a major technological shift while emphasizing the AI safety.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Cybersecurity sections to stay informed and up-to-date with our daily articles.