The Leak That Shook the AI World

On March 26, 2026, the AI industry got an unexpected peek behind Anthropic's curtain. A human configuration error in the company's content management system left a publicly searchable data cache containing nearly 3,000 unpublished assets — including what appeared to be a full draft blog post announcing a new model tier the company calls Claude Mythos.

The documents were independently discovered by two security researchers: Roy Paz, a senior AI security researcher at LayerX Security, and Alexandre Pauwels, a cybersecurity researcher at the University of Cambridge. Fortune broke the story after reviewing the leaked materials.

What Is Claude Mythos?

According to the leaked draft, Mythos sits above Anthropic's current top tier — Opus — in a new model family codenamed Capybara. The document describes it plainly:

"'Capybara' is a new name for a new tier of model: larger and more intelligent than our Opus models — which were, until now, our most powerful."

Anthropic confirmed the model's existence after being contacted by Fortune, calling it "a step change" and "the most capable we've built to date." In an official statement, the company said:

"We're developing a general purpose model with meaningful advances in reasoning, coding, and cybersecurity. Given the strength of its capabilities, we're being deliberate about how we release it."

The model is currently being tested with a small group of early-access enterprise customers. It's described as expensive to run and not yet ready for general release — Anthropic is working on making it more efficient before any public launch.

The Cybersecurity Alarm

The most alarming element of the leak wasn't that a powerful new model exists — it's how Anthropic describes its risks. The draft document claims Mythos is "currently far ahead of any other AI model in cyber capabilities" and warns that it "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of cyber defenders."

Anthropic was reportedly already briefing top government officials in private, warning that Mythos could make large-scale cyberattacks significantly more likely in 2026, according to Axios. Cybersecurity stocks slumped in the days following the leak as the implications sank in.

The concern is structural: as AI agents become more autonomous and capable of multi-step reasoning, hackers can run simultaneous, coordinated attacks that scale beyond anything human defenders can match in real time. Mythos — even before its public release — appears to be a meaningful step in that direction.

Capybara vs. Mythos: What's in a Name?

The leaked documents use both names somewhat interchangeably. "Capybara" appears to be the product tier name, while "Mythos" is the internal model name. Think of it like how "Claude 3" was a tier and "Opus" was a specific model within it.

Anthropic says it "completed training" on Mythos and is in the early access phase. The blog post describes it as:

"By far the most powerful AI model we've ever developed."

Compared to Claude Opus 4.6 (the current flagship), the draft post says Mythos "gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others."

An Uncomfortable Irony

There's a certain irony in a company warning about a model's unprecedented cybersecurity risks while simultaneously leaking it through a basic configuration mistake. Anthropic attributed the exposure to "human error in the configuration of its content management system" and removed public access to the data store after being notified by Fortune.

The episode raises real questions about internal security practices at frontier AI labs — especially as the models they're building become more dangerous in the wrong hands. If an AI company can accidentally expose its most sensitive product roadmap, what does that suggest about the security posture of the systems themselves?

What Comes Next

Anthropic hasn't announced a public release date. The model remains in controlled early-access testing, with the company emphasizing caution. Given its stated cybersecurity implications, expect significant safety evaluations and likely some form of government briefing before any wide rollout.

For now, Mythos is the most powerful AI model most people have never been allowed to use — and the world just found out it exists by accident.