Anthropic's Most Powerful AI Yet, Claude Mythos, Exposed in Massive Data Leak

Not a day goes by without news about Anthropic, currently the hottest AI company in the world: An accidentally publicly accessible data store has revealed internal materials from AI company Anthropic, including drafts of blog posts and unpublished content surrounding a new AI model that has not yet been officially announced. The model is internally named "Claude Mythos" and is said, according to the company, to be the most powerful system Anthropic has ever developed. The leak immediately triggered strong movements in the financial markets.

According to the leaked documents, "Mythos" represents a new model category that goes beyond Anthropic's previous top tier. Until now, the company offered three tiers: Haiku, Sonnet, and Opus. Mythos is set to form a fourth, even more powerful level.

According to the internal materials, the name was deliberately chosen to evoke "the deep connective tissue that links knowledge and ideas together." Compared to the previous flagship model Claude Opus 4.6, Mythos achieves significantly higher scores in the areas of software development, academic reasoning, and cybersecurity.

"Mythos is currently far ahead of any other AI model in cyber capabilities and heralds an imminent wave of models that can exploit vulnerabilities in ways that far exceed the efforts of defenders," the leaked documents state.

Should the new model really come to market soon and prove better than Claude Opus 4.6, that would be a big deal. After all, Claude Opus 4.6 is currently at the top of the AI charts, meaning Anthropic would be widening its lead over rivals Google, OpenAI, and others.

A central theme of the leaked documents is the model's exceptional performance in the field of cybersecurity. The model is said to be capable of quickly identifying and exploiting vulnerabilities in software systems. Anthropic itself warns in the internal documents of the significant risks that come with this.

For this reason, the company is planning a particularly cautious and gradual market launch. In a first phase, selected organizations from the cyber defense sector will receive early access in order to strengthen their systems against the coming wave of AI-powered attacks. Anthropic emphasizes that Mythos is also very costly to operate due to its size and computational intensity.

The leak immediately triggered strong price movements in technology and cybersecurity companies. The fear that such a powerful model could render existing security solutions obsolete drove investors to sell.

Stocks of cybersecurity companies such as Palo Alto Networks (PANW), CrowdStrike (CRWD), and Fortinet (FTNT) fell sharply by 4 to 6 percent, and the iShares Expanded Tech-Software ETF (IGV) also dropped significantly.

According to the business magazine Fortune, around 3,000 files from Anthropic's blog infrastructure were accidentally stored in a publicly accessible data store. These included drafts for announcements as well as internal material that had not yet been intended for publication. Anthropic has not yet officially commented on the incident.

Also not to be overlooked: Anthropic is currently in the fast lane and has made significant inroads particularly in the B2B and coding market as well as in office software. The company also plans to go public this year -- and buzz around new AI models is always a welcome boost. As recently reported, Claude is currently suffering from massive outages. These could be the result of a flood of users turning to Anthropic.