Anthropic, the US-based artificial intelligence company, has confirmed the existence of a powerful new AI model, “Claude Mythos,” after a human error in its content management system accidentally exposed a draft blog about the technology. The company said the model represents its most advanced system yet, with capabilities that could have far-reaching implications for cybersecurity.
The leak occurred last week when a draft blog, stored in an unsecured and publicly searchable location, became accessible online. The blog described Mythos as part of a new tier of AI models, codenamed Capybara, which is larger and more capable than Anthropic’s previous Opus models. The exposed document warned that the model “poses unprecedented cybersecurity risks.”
An Anthropic spokesperson told Fortune that Mythos represents a “step change” in AI development, with significant advances in reasoning, coding, and cybersecurity. The spokesperson described it as the most capable model the company has built to date.
Experts have raised concerns that the technology could amplify risks in the cybersecurity landscape. AI systems like Mythos are capable of acting autonomously and tackling complex tasks with precision, making them potentially attractive tools for hackers. According to reports, Anthropic has privately briefed top government officials, cautioning that Mythos could make large-scale cyberattacks more likely in 2026. Following news of the leak, cybersecurity stocks reportedly fell as investors weighed the implications of a more capable AI in malicious hands.
The model is extremely compute-intensive and expensive to operate, and Anthropic said it is working to improve its efficiency before any public release. While exact timelines for availability have not been announced, the company emphasized ongoing development to ensure both functionality and safety.
Industry observers note that advanced AI agents, which learn to act without human input, could be exploited to conduct multiple simultaneous cyberattacks. Employees using AI in their work systems may inadvertently create vulnerabilities, increasing the risk of breaches. AI’s ability to analyze and imitate human behavior could also make it easier for attackers to compromise personal and corporate identities, according to a Palo Alto Networks executive.
Mythos and its Capybara tier of models appear to mark a new phase in AI development, with autonomous reasoning, coding, and problem-solving capabilities beyond Anthropic’s previous systems. The company’s caution in rollout reflects awareness of the model’s potential impact on digital security.
Euronews Next has reached out to Anthropic for additional comment, as experts continue to monitor the implications of Mythos for cybersecurity and AI governance worldwide.
This leak highlights the growing intersection of cutting-edge AI and digital security, underscoring the need for careful oversight as AI models become increasingly sophisticated.