# Claude Mythos

> **TL;DR.** [Claude Mythos](https://red.

- **Category:** AI / Frontier Models / AI Safety
- **Stage:** validating
- **Age:** 57 days
- **Origin date:** 2026-04-07
- **First detected:** 2026-04-15
- **Canonical URL:** https://earlyterms.com/term/claude-mythos
- **Sources:** 8 primary URLs

## Definition

[Claude Mythos](https://red.anthropic.com/2026/mythos-preview/) is an unreleased frontier LLM from Anthropic, publicly previewed on April 7, 2026. Positioned as general-purpose, the preview centres on cybersecurity: Anthropic used it to identify thousands of candidate zero-days across every major OS and browser, including a 27-year-old OpenBSD TCP SACK integer-overflow found on a single scaffold run.

The defining feature is Anthropic stating it will not ship the model. Instead, it launched [Project Glasswing](https://www.anthropic.com/glasswing) — a coalition with AWS, Apple, Google, Microsoft, Linux Foundation, JPMorganChase, CrowdStrike, and 40+ critical-software maintainers — for gated defensive access. Mythos is framed as proof that weights-level safeguards must arrive before any Mythos-class model ships at scale.

## Analogy

Anthropic publishing a recall notice on a car it decided never to sell — and using the recall itself as marketing for the next model.

## Why it's emerging now

On April 7, 2026 Anthropic became the first frontier lab to publicly name a model it built and refused to ship. The preview's concrete capability — a 27-year-old OpenBSD kernel bug found on a single scaffold run for under $50 — plus the Project Glasswing coalition of AWS, Apple, Google, Microsoft, Linux Foundation and JPMorganChase locked the term into AI-policy coverage. On April 16 Anthropic reinvoked 'Mythos' at the Opus 4.7 launch.

## Related terms

- *alias:* Mythos
- *parent:* Claude
- *related:* Project Glasswing
- *competitor:* Claude Opus 4.7
- *competitor:* Claude Opus 4.6
- *competitor:* GPT-5.4
- *competitor:* Gemini 3.1 Pro
- *related:* responsible scaling policy
- *related:* ASL-4
- *related:* AI red-teaming
- *related:* zero-day vulnerability
- *related:* Cal.com closed source

## Sources

1. [Anthropic — Claude Mythos Preview (red.anthropic.com)](https://red.anthropic.com/2026/mythos-preview/)
2. [Anthropic — Project Glasswing](https://www.anthropic.com/glasswing)
3. [UK AISI — Evaluation of Claude Mythos Preview's cyber capabilities](https://www.aisi.gov.uk/blog/our-evaluation-of-claude-mythos-previews-cyber-capabilities)
4. [Help Net Security — Anthropic's new AI model finds zero-days across every major OS and browser](https://www.helpnetsecurity.com/2026/04/08/anthropic-claude-mythos-preview-identify-vulnerabilities/)
5. [VentureBeat — Mythos autonomously exploited vulnerabilities that survived 27 years](https://venturebeat.com/security/mythos-detection-ceiling-security-teams-new-playbook)
6. [Simon Willison — Project Glasswing sounds necessary](https://simonwillison.net/2026/Apr/7/project-glasswing/)
7. [Tom's Hardware — Claude Mythos isn't a sentient super-hacker, it's a sales pitch](https://www.tomshardware.com/tech-industry/artificial-intelligence/anthropics-claude-mythos-isnt-a-sentient-super-hacker-its-a-sales-pitch-claims-of-thousands-of-severe-zero-days-rely-on-just-198-manual-reviews)
8. [Hacker News — System Card: Claude Mythos Preview (846 points)](https://news.ycombinator.com/item?id=47679258)

---
_Generated by EarlyTerms · https://earlyterms.com/term/claude-mythos_