Anthropic Claude Mythos Model Leaked in Data Store — A Step Change With Unprecedented Cybersecurity Risks
Anthropic has confirmed it is testing a new AI model called Claude Mythos that represents a "step change" in capabilities — after a draft blog post was accidentally leaked from an unsecured, publicly searchable data store. The leak revealed that Anthropic believes the model poses "unprecedented cybersecurity risks," making this one of the most consequential accidental disclosures in AI history.
What Was Leaked
Around 3,000 assets linked to Anthropic's blog were found in a publicly accessible data store, including draft announcements and internal content. Among them was a blog post identifying a new model called Claude Mythos — internally codenamed "Capybara" — described as a tier above the current Opus models.
The leaked documents reveal that compared to Claude Opus 4.6, Mythos achieves dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity. The model is positioned as larger, more intelligent, and more expensive than anything Anthropic has released to date.
Anthropic's Response
After the leak went public, Anthropic confirmed the model's existence, stating that Claude Mythos is "the most capable we've built to date" and represents "a step change" in AI performance. The company said it is currently being trialed by early access customers.
Anthropic attributed the exposure to "a configuration error in one of its external content management tools," calling it "human error." The irony hasn't been lost on anyone — a company that prides itself on AI safety accidentally exposed its most sensitive product roadmap to the public internet.
The Cybersecurity Concern
Perhaps the most alarming detail from the leak is Anthropic's own assessment that Mythos poses "unprecedented cybersecurity risks." This suggests the model's coding and reasoning capabilities are powerful enough to potentially be weaponized for sophisticated cyberattacks — a scenario Anthropic's safety team has been working to mitigate before public release.
This disclosure adds fuel to the ongoing debate about whether AI companies should develop models they themselves consider dangerous, even with safety measures in place.
What "Capybara" Tier Means
The leaked documents describe Capybara as a new model tier — above the existing Haiku (fast/cheap), Sonnet (balanced), and Opus (powerful) tiers. This suggests Anthropic is creating an ultra-premium tier for its most capable models, likely aimed at enterprise customers and researchers willing to pay significantly more for cutting-edge performance.
Bottom Line
Anthropic accidentally revealing its most powerful AI model through a misconfigured data store is both embarrassing and significant. The Claude Mythos leak shows that the next generation of AI is coming fast — and that even the company building it considers the cybersecurity implications unprecedented. Whether this accelerates or delays the official launch remains to be seen, but the AI industry now knows what Anthropic has been building behind closed doors.
Frequently Asked Questions
What is Claude Mythos?
Claude Mythos is Anthropic's next-generation AI model, internally codenamed "Capybara." It sits above the current Opus tier and delivers dramatically higher performance in coding, reasoning, and cybersecurity benchmarks.
When will Claude Mythos be released?
No official release date has been announced. Anthropic says the model is currently being tested by early access customers. The leak may impact the timeline.
What are the cybersecurity risks?
Anthropic's own leaked assessment describes "unprecedented cybersecurity risks," suggesting the model's capabilities in code generation and reasoning could potentially be exploited for sophisticated cyberattacks if not properly safeguarded.
How did the leak happen?
A configuration error in one of Anthropic's external content management tools left approximately 3,000 blog-related assets — including draft announcements — publicly accessible and searchable online.