What is Claude Mythos? Anthropic's Most Powerful AI Model Explained

8 min read
  • claude mythos
  • anthropic
  • claude mythos model
  • what is claude mythos
  • project glasswing

Claude Mythos is the most powerful AI model Anthropic has ever built. It is not available to the public. It found thousands of previously unknown security vulnerabilities in the world's software infrastructure. And its existence was first revealed by accident — when Anthropic left a draft blog post about it in an unsecured, publicly searchable data cache.

This is the story of Claude Mythos — what it is, how it got here, what it can actually do, and why it has sent banks, governments, and tech companies scrambling.

How Claude Mythos Was Discovered

On March 26, 2026, Fortune reported that Anthropic had accidentally exposed a pre-release draft blog post about a new model in an unsecured, publicly accessible data store. The post was reviewed by Roy Paz, a senior AI security researcher, who alerted Fortune to its existence.

The draft described a model called Claude Mythos — codenamed internally as "Capybara" — and described it as "by far the most powerful AI model we've ever developed." The document also described Mythos as posing unprecedented cybersecurity risks.

Anthropic confirmed the model's existence in response to questions. A company spokesperson said Mythos represents "a step change" in AI performance and is "the most capable we've built to date," adding that it was being tested with a small group of early access customers.

Claude Mythos Preview was officially released on April 8, 2026 — not to the general public, but to a select group of companies as part of a security initiative called Project Glasswing.

What Is Claude Mythos?

Claude Mythos is Anthropic's newest and most powerful general-purpose AI model. It sits in a new tier above Opus — the company's previous most powerful model family.

In Anthropic's own words: "Mythos 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."

The name Mythos was chosen deliberately. Anthropic describes it as evoking "the deep connective tissue that links together knowledge and ideas."

The benchmarks confirm a generational leap over previous models. Claude Mythos Preview scored 93.9% on SWE-bench — the industry standard for real-world software engineering tasks — and 97.6% on USAMO, a measure of advanced mathematical reasoning. Anthropic's own materials describe these results as "far ahead of any other AI model."

What Makes It Different — The Cybersecurity Capability

While Mythos performs strongly across all general tasks, its cybersecurity capability is what set off alarm bells across the industry.

Mythos can autonomously find, analyze, and in some cases exploit software vulnerabilities — at a scale and speed no previous AI model could match. In Anthropic's own testing, Mythos discovered thousands of previously unknown vulnerabilities — called zero-day vulnerabilities — in major software systems including most major operating systems.

In one specific test across approximately 7,000 entry points into software repositories, Mythos achieved 595 crashes at tiers 1 and 2, and achieved full control flow hijack on ten separate, fully patched targets — the highest tier of exploitation difficulty.

The UK Government's AI Security Institute independently evaluated Claude Mythos Preview and confirmed that while it was not significantly better at individual cybersecurity tasks compared to other models, it completed difficult multistep infiltration challenges with a high level of frequency — including some that no other AI model had completed before.

This is the capability that sent global banks, tech giants, and governments scrambling. Anthropic CEO Dario Amodei described the risk plainly: "The danger is just some enormous increase in the amount of vulnerabilities, in the amount of breaches, in the financial damage that's done from ransomware on schools, hospitals, not to mention banks."

What Is Project Glasswing?

Project Glasswing is Anthropic's response to its own model's capabilities. Recognizing that Mythos could be weaponized as easily as it could be used for defense, Anthropic chose not to release it publicly. Instead it launched Project Glasswing — a controlled, invite-only program that gives a small number of critical industry partners and open source developers access to Mythos specifically to help them find and fix vulnerabilities in their own systems before bad actors can exploit them.

The idea: use the most powerful offensive AI to build defenses, before the capability spreads more broadly.

Companies confirmed to have received access through Project Glasswing include Apple, Amazon, JPMorgan Chase, and Palo Alto Networks. Claude Mythos Preview is also available in Private Preview on Google Cloud's Vertex AI platform.

Anthropic has stated it does not plan to make Claude Mythos Preview generally available, but says its goal is to learn from real-world deployment of safeguards to work toward an eventual broad release of Mythos-class models.

The Controversy — Is the Threat Real?

Not everyone accepted Anthropic's framing at face value. Several prominent cybersecurity researchers pushed back on what some called a coordinated hype campaign.

Renowned security researcher Bruce Schneier said Anthropic was "convincing a lot of people that Mythos is this amazing step change in capability when the evidence right now… is that it might not be."

Cybersecurity firm Aisle found that many of Mythos's headline results could be reproduced using cheaper models working in parallel. Aisle founder Stanislav Fort wrote: "A thousand adequate detectives searching everywhere will find more bugs than one brilliant detective who has to guess where to look." The argument being that scale and coordination matter more than having the newest model.

Ben Harris, CEO of cybersecurity firm watchTowr, told CNBC: "What we are seeing across the industry now is that people are able to reproduce the vulnerabilities found with Mythos through clever orchestration of public models to get very, very similar results."

Anthropic did not dispute that earlier models were capable of finding software vulnerabilities. The company had previously shown that Claude Opus 4.6 — a widely available model — had already found more than 500 high severity vulnerabilities in open-source software.

The honest read: Mythos is genuinely more capable than its predecessors, and the cybersecurity threat from AI is real. Whether Mythos specifically represents an unprecedented leap or a meaningful but incremental step is a legitimate debate. What is not debatable is that AI-enabled vulnerability discovery is accelerating — and the gap between finding vulnerabilities and patching them is widening.

The Industry Response

The controlled rollout of Mythos triggered immediate responses across industry and government.

OpenAI announced GPT-5.5-Cyber — a model specifically tailored for cybersecurity — within weeks of Mythos's release, allowing limited access to vetted cybersecurity teams.

The Trump administration began considering new government oversight over future models in direct response to Mythos's capabilities.

High-profile meetings between Trump administration officials, tech CEOs, and bank chief executives were convened specifically to discuss the security risks raised by Mythos-class models.

Claude Opus 4.7 — The Safer Alternative

On April 16, 2026, Anthropic released Claude Opus 4.7 — described as an improvement over previous models but "less broadly capable" than Mythos Preview. Specifically, its cyber capabilities are not as advanced.

Claude Opus 4.7 is available publicly across all of Anthropic's Claude products, its API, and through cloud providers Microsoft, Google, and Amazon at the same price as Claude Opus 4.6.

The positioning is deliberate: Opus 4.7 gives developers and businesses access to Anthropic's best generally available model, while Mythos remains restricted to Project Glasswing partners until Anthropic determines how to release it safely at scale.

Can You Access Claude Mythos?

Not directly — unless you are one of the select Project Glasswing partners. Anthropic has not announced a timeline for general availability.

If you want the most capable publicly available Claude model right now, that is Claude Opus 4.7, available at claude.ai and via the Anthropic API using the model string claude-opus-4-7.

For developers on Google Cloud, Claude Mythos Preview is available in Private Preview on Vertex AI — but access requires applying through Google Cloud's enterprise channels.

The Bottom Line

Claude Mythos is the most significant AI model release of 2026 — not because of its benchmark scores, but because of what it represents. It is the first AI model whose capabilities were judged too dangerous for public release by its own creators. That is a meaningful milestone regardless of where you stand on the hype debate.

Whether Anthropic's caution is genuinely unprecedented responsibility or sophisticated marketing — or both — the questions it has raised about who should control the most powerful AI systems, and how, are not going away.

The Neuron covers AI clearly — no hype, no jargon. This article was last updated May 8, 2026, based on the latest reporting from CNBC, Fortune, and Anthropic's official blog.