Skip to main content
← Back to Blog

What Is Claude Mythos? Who Can Use It and Why It Matters

Brandon Veiseh, Co-Founder & CEO at MindFort

Written by

Brandon Veiseh

AI Research2026-04-08·Updated 2026-10-30·9 min read

Claude Mythos is Anthropic's most capable family of models for finding and exploiting software vulnerabilities. Mythos 5.1 is the current version and launched on September 1, 2026. It is the same model as the generally available Fable 5.1 with looser cyber safeguards. Only vetted organizations can use it through Anthropic's Cyber Verification Program.

Claude Mythos is the model that made security teams rethink how fast bugs get found. Anthropic has kept it gated since April 2026 because of its offensive cyber skills. This guide covers what Mythos is and which versions exist. It also covers who can use it today and what it means for your applications.

What is Claude Mythos?

Claude Mythos is Anthropic's most capable model family, and its standout skill is security work. Anthropic disclosed Claude Mythos Preview  on April 7, 2026. In internal testing it found thousands of high-severity zero-days across every major operating system and web browser. Anthropic says these skills "emerged as a downstream consequence" of better coding, reasoning, and autonomy.

Some of the bugs had survived decades of review. One was a 27-year-old bug in OpenBSD  that cost about $50 in compute to find. Another was CVE-2026-4747 , a 17-year-old FreeBSD remote code execution flaw that Mythos exploited on its own. On a Firefox JavaScript engine benchmark, Opus 4.6 produced 2 working exploits  in several hundred attempts. Mythos Preview produced 181.

Which versions of Claude Mythos exist?

There have been three Mythos releases so far. Each newer version shipped alongside a public sibling model with tighter safeguards.

VersionReleasedWho can use itPublic sibling
Mythos PreviewApril 7, 2026Project Glasswing partnersNone
Mythos 5 June 9, 2026Project Glasswing partnersFable 5
Mythos 5.1 September 1, 2026Verified organizations in trusted access programsFable 5.1

Access has not been steady. A US export control directive  suspended Fable 5 and Mythos 5 for every customer on June 12. Anthropic restored access on July 1  after adding a new safeguard classifier.

Can you use Claude Mythos today?

Only if your organization is verified. The Mythos 5.1 docs  say the model is open only to organizations verified through Anthropic's programs. On October 6 the Cyber Verification Program  added Mythos 5.1 to all three of its tiers:

TierWho it is forReview time
Defense AccessDefensive work by teams, open-source maintainers, and individual researchersA few days
Red Team AccessAuthorized pentesting by organizationsA few weeks
Specialized AccessA small set of organizations reviewed with the US governmentIn-depth review

Verified teams can call Mythos 5.1 through the Claude API and Google Cloud. It also runs on Microsoft Foundry and eligible Amazon Bedrock accounts. Anthropic lists it at $10 per million input tokens and $50 per million output tokens . Use also requires a 30-day data retention policy for safety monitoring. If you can't get access, see our guide to Claude Mythos alternatives for pentesting.

How is Claude Mythos different from Fable 5.1?

Anthropic says Fable 5.1 and Mythos 5.1 are the same model  with different safeguards. Fable 5.1 is generally available on every major cloud. It can spot vulnerabilities in source code. It blocks penetration testing and exploit generation , along with binary vulnerability scanning.

Mythos 5.1 relaxes those cyber blocks for vetted users. That makes it the version that can attack a running target. For a closer look at what the public model can do, read our Fable 5 security review.

What is Project Glasswing?

Project Glasswing is how Anthropic first shared Mythos with defenders. It launched in April  with twelve partners. They included AWS, Apple, and Microsoft. In June Anthropic added about 150 more organizations  across more than 15 countries. It also backed the program with $100 million in usage credits.

In October Glasswing members moved into the Specialized Access tier of the Cyber Verification Program. Anthropic says partners uncovered at least 129,000 verified vulnerabilities  between April and July 2026. More than 33,000 of them were rated critical or high.

Why is Claude Mythos controversial?

Governments treat Mythos-class models as a national security issue. The June suspension followed a report that Amazon researchers had found a way around Fable 5's safeguards . Anthropic complied with the directive. It also argued that one narrow jailbreak should not recall a commercial model.

Testing has raised questions too. In July Anthropic reviewed 141,006 cyber evaluation runs  and found three real-world incidents. In one, a Mythos 5 run published a malicious PyPI package while a misconfigured test machine had live internet access. Anthropic called it "closer to a harness and operational failure" than an alignment failure.

Why should security teams care about Claude Mythos?

The cost of finding a serious bug has collapsed. Work that took a skilled researcher weeks now takes hours and a few dollars of compute. Anthropic expects many other AI companies to have Mythos-class models  within 6 to 12 months. Some may ship them without comparable safeguards.

Most application security programs assume that exploiting a bug is hard, slow, and expensive. That assumption no longer holds. Every legacy codebase, every open-source dependency, every piece of internal tooling is now exposed in a way it has never been before.

How can you protect your applications from Mythos-class attackers?

Test your applications as often as you ship them. MindFort deploys autonomous offensive security swarms that pen test your applications around the clock. They probe authentication flows, business logic, and APIs. These are the same bug classes Mythos proved AI can now find at scale.

Each confirmed finding comes with a working exploit. MindFort then opens a pull request with the smallest code change that fixes the issue, plus a threat model that explains it. The agents retest the fix, so you know the hole is closed. Plans start at $499 a month.

What should you do right now?

Stop treating the annual pen test as your main defense. Attackers will get Mythos-class tools on their own timeline. Your testing has to keep pace with your release cycle.

Start with your most exposed applications and test them continuously. Fix what is exploitable first. The only question is whether you find the bugs first, or an attacker does.

See how autonomous security agents can protect your applications → 

FAQ

What is Claude Mythos?

Claude Mythos is Anthropic's most capable model family for cybersecurity work. It can find and exploit software vulnerabilities on its own. Anthropic first disclosed it as Mythos Preview in April 2026. The current version is Mythos 5.1.

Can I use Claude Mythos?

Only if your organization is verified. Mythos 5.1 is available through Anthropic's trusted access programs, such as the Cyber Verification Program. There is no self-serve sign-up. Most teams can use Fable 5.1 instead. It is generally available but blocks pentesting and exploit generation.

Are Claude and Claude Mythos different?

Claude is the name of Anthropic's whole model lineup. Mythos is one model family inside it. Fable 5.1 and Mythos 5.1 are the same model with different safeguards. Mythos 5.1 allows more offensive security work for vetted users.

Why is Claude Mythos controversial?

Its offensive cyber skills worry governments and researchers. On June 12, 2026, a US export control directive forced Anthropic to suspend Fable 5 and Mythos 5. Access returned on July 1. Anthropic later disclosed that a Mythos 5 run published a malicious PyPI package during a misconfigured evaluation.

What is Project Glasswing?

Project Glasswing is Anthropic's program that gave vetted defenders early Mythos access for defensive security work. It launched in April 2026 with twelve major technology and finance companies. It grew to about 200 organizations by June. In October its members moved into the top tier of the Cyber Verification Program.

How can organizations protect themselves against AI-powered vulnerability discovery?

Test continuously at the pace you ship. An annual pen test leaves most of the year uncovered. MindFort runs autonomous security agents that pen test your applications around the clock. Each finding comes with a proven exploit and a fix as a pull request.

About the author

Brandon Veiseh, Co-Founder & CEO at MindFort

Brandon Veiseh

Co-Founder & CEO · MindFort

Founded his first startup building NLP models for network packet inspection. Led product at ProjectDiscovery, built their enterprise platform from scratch. At NetSPI, led development of AI tools for offensive security.

An Autonomous Security Agent.

Agents find vulnerabilities and fix them for you.

Book a demo with our team.

First Results

Hours

Coverage

24/7

False Positives

<1%

Setup

Minutes