Exclusive: Anthropic Confirms New AI Model with Major Advancements After Data Leak

Anthropic Unveils New AI Model in Early Testing Phase

Anthropic, a leading artificial intelligence company, is currently developing and testing a new AI model that surpasses any of its previous releases. This revelation comes after a data leak exposed details about the model’s existence. The company has confirmed that the new model represents a significant advancement in AI performance and is considered the most capable it has built to date.

The information about the model was inadvertently stored in a publicly-accessible data cache, which was later reviewed by Jendela Magazine. A draft blog post found in this cache revealed the model’s name as “Claude Mythos” and highlighted concerns about potential cybersecurity risks associated with it.

In addition to the model details, the data cache also contained information about an upcoming CEO summit in Europe, which is part of Anthropic’s strategy to attract large corporate customers. According to cybersecurity researchers, the unsecured data lake included various documents, some of which were early drafts of content that had not been published on the company’s official sites.

Details of the Data Leak

A senior AI security researcher at LayerX Security, Roy Paz, and Alexandre Pauwels from the University of Cambridge reviewed the leaked documents. They noted that there were approximately 3,000 assets related to Anthropic’s blog that were accessible to the public but had not been published elsewhere. These materials were shared with Jendela Magazine for assessment.

After being informed of the data leak, Anthropic took steps to secure the data store and prevent further access. In a statement provided to Jendela Magazine, the company acknowledged a “human error” in the configuration of its content management system, which led to the exposure of the draft blog post.

The draft blog post also mentioned a new tier of AI models called “Capybara.” According to the document, Capybara is a larger and more intelligent model than the Opus series, which were previously the most powerful. However, Capybara is expected to be more expensive compared to the existing models.

Cybersecurity Concerns

The leaked document raised significant concerns about the cybersecurity risks posed by the new AI model. It stated that the model could potentially exploit vulnerabilities in ways that outpace the efforts of defenders. Anthropic emphasized the need for caution and plans to release the model to cyber defenders first, allowing them to prepare for future AI-driven exploits.

This concern aligns with similar issues faced by other AI companies. For instance, OpenAI recently classified GPT-5.3-Codex as a high-capability model for cybersecurity tasks. Anthropic’s Opus 4.6 model also demonstrated the ability to identify software vulnerabilities, highlighting the dual-use nature of such technologies.

Moreover, Anthropic has reported attempts by hacking groups, including those linked to the Chinese government, to exploit its AI systems. One case involved a state-sponsored group using Claude Code to infiltrate multiple organizations before being detected.

The Exclusive Executive Retreat

The leak of confidential information appears to stem from an error within the company’s content management system (CMS). Digital assets created using the CMS are typically set to public by default, leading to the accidental publication of a large cache of images, PDF files, and audio files.

An Anthropic spokesperson attributed the issue to “human error” and confirmed that the problem was related to an external CMS tool. Many of the documents appeared to be discarded or unused assets, while others were meant to be private or internal. One document even referenced an employee’s “parental leave.”

The leak also revealed details about an invite-only retreat for European CEOs, hosted by Anthropic CEO Dario Amodei. The event, described as an “intimate gathering,” aims to facilitate discussions on the future of AI and provide attendees with exclusive insights into unreleased Claude capabilities.