Anthropic Data Leak Exposes Highly Advanced Artificial Intelligence Model
San Francisco, Friday, 27 March 2026.
An accidental data leak at Anthropic has revealed “Claude Mythos,” an unreleased artificial intelligence model that reportedly possesses unprecedented capabilities and poses significant new cybersecurity risks.
A Security Lapse Exposes the Future of AI
The inadvertent disclosure stemmed from a misconfiguration in one of Anthropic’s external content management systems (CMS) [3]. A default setting caused approximately 3,000 previously unpublished assets to become publicly searchable [1][3]. Cybersecurity researchers Roy Paz and Alexandre Pauwels independently discovered the unsecured data cache, which contained draft blog posts, images, audio files, and sensitive internal documents, including an employee’s parental leave details [1][3]. Anthropic secured the database shortly after being notified by Fortune in late March 2026 [1][3].
Enter Claude Mythos and the Capybara Tier
The centerpiece of the leaked documents is “Claude Mythos,” a model currently undergoing testing with select enterprise customers [1][2]. Mythos is slated to introduce a new premium tier designated as “Capybara” [1]. Internal assessments describe the Capybara models as significantly larger and more intelligent than the existing “Opus” tier, demonstrating dramatically higher performance in software coding, academic reasoning, and cybersecurity evaluations [1]. An Anthropic spokesperson confirmed the model’s existence, calling it a “step change” and “the most capable we’ve built to date” [1][2].
Strategic Moves and Executive Gatherings
Beyond technical specifications, the data leak also uncovered Anthropic’s strategic business maneuvers [1]. The documents detailed plans for an exclusive, invite-only retreat for chief executive officers of major European companies [1][3]. Scheduled to take place at an 18th-century manor-turned-hotel-and-spa in the English countryside, the two-day summit will feature Anthropic CEO Dario Amodei [1][3]. The event aims to facilitate discussions on enterprise AI adoption and showcase unreleased Claude capabilities, though it remains unclear if the leak will alter the summit’s schedule [alert! ‘The leaked documents do not confirm if the summit timeline will be adjusted following the public exposure of the event’]. Anthropic confirmed the event is part of an ongoing series designed to engage European business leaders on the future of artificial intelligence [1].