---
title: 'Claude Mythos'
description: "Anthropic's Claude model line deployed through trusted access programs with safeguards designed for advanced cybersecurity and life-sciences work."
canonical_url: 'https://darkfactory.dev/glossary/mythos'
markdown_url: 'https://darkfactory.dev/glossary/mythos.md'
collection: glossary
date_published: '2026-10-07T00:00:00-04:00'
date_modified: '2026-10-07T00:00:00-04:00'
---

# Claude Mythos


## Definition

Claude Mythos is Anthropic's Claude model line deployed through trusted access programs with safeguards designed for advanced cybersecurity and life-sciences work. It is a general-purpose language model line whose access policy reflects capabilities that Anthropic considers unusually sensitive in those domains.

For the 5 and 5.1 releases, Anthropic states that Mythos and Fable share the same underlying model. Their safeguards and access rules differ. Mythos does not mean a model with every safeguard removed.

## Origin and attribution

Anthropic introduced Claude Mythos Preview on April 7, 2026 with Project Glasswing, a program to apply the model to defensive security work. Nicholas Carlini and coauthors' technical report described the company's cybersecurity evaluations. The report's author list does not establish those researchers as the model's sole creators.

Claude Mythos 5 followed on June 9, 2026, alongside the generally available Fable 5. Claude Mythos 5.1 and Fable 5.1 launched on September 1, 2026. Mythos Preview, Mythos 5, and Mythos 5.1 are distinct releases, so results from one should not be transferred to another without evaluation.

## Scope and limits

Mythos names a branded model line and deployment policy. Mythos-class is Anthropic's capability label; it is not an independently standardized threshold for intelligence or safety. Capabilities need evaluation on the intended task even when model access is restricted.

Access to Mythos 5.1 depends on verification through Anthropic's trusted access programs. The permissions and safeguards differ by use case. That provider approval does not authorize a customer to test an unrelated system, access private data, or bypass the controls in its own organization.

## Operational significance

Security and research workflows need an approved scope, isolated execution where appropriate, and verification of each finding. A model-generated vulnerability report needs reproduction and triage; a research hypothesis needs supporting evidence and experimental checks.

Evaluate the exact deployment. Fable and Mythos can receive different scores when Fable's safeguards intervene, even though the underlying model is shared. Those scores measure the deployed system's behavior, including its safeguards.

## Distinguish it from nearby terms

- Fable is the generally available deployment of the shared 5.1 model with additional safeguards. Mythos uses trusted access and safeguards suited to approved specialist work.
- Claude is the parent family. Mythos is one named line and deployment within it.
- Project Glasswing is a defensive-security program, not a model version or a synonym for Mythos.
- A verification gate checks whether an action or artifact may proceed. Access to a model does not replace that gate.

## Check your understanding

A vetted security team receives Mythos access and a convincing report about a third-party service. What further authorization and evidence are required before testing the service or reporting a verified vulnerability?

## Also called

Mythos

## Related terms

- [Claude](https://darkfactory.dev/glossary/claude)
- [Claude Fable](https://darkfactory.dev/glossary/fable)
- [Large language model (LLM)](https://darkfactory.dev/glossary/large-language-model)
- [Reasoning model](https://darkfactory.dev/glossary/reasoning-model)
- [Evaluation (eval)](https://darkfactory.dev/glossary/evaluation)
- [Verification gate](https://darkfactory.dev/glossary/verification-gate)
- [Prompt injection](https://darkfactory.dev/glossary/prompt-injection)

## Related factory areas

- [Security, privacy & software supply chain](https://darkfactory.dev/factory/security)
- [Governance, audit, legal & accountability](https://darkfactory.dev/factory/governance-accountability)
- [Verification, evaluation & quality truth](https://darkfactory.dev/factory/verification)

## Evidence and further reading

- [Anthropic: Assessing Claude Mythos Preview's cybersecurity capabilities](https://www.anthropic.com/research/mythos-preview)
- [Anthropic: Claude Fable 5 and Claude Mythos 5](https://www.anthropic.com/news/claude-fable-5-mythos-5)
- [Anthropic: Claude Fable 5.1 and Claude Mythos 5.1](https://www.anthropic.com/claude-fable-and-mythos-5-1)
- [Anthropic: Claude Mythos](https://www.anthropic.com/claude/mythos)