Anthropic Is About to Release Mythos to the Public — The AI That Found 10,000 Zero-Days in a Month

Breaking — Anthropic is preparing to release a public version of Claude Mythos — the cybersecurity AI model that autonomously discovered zero-day vulnerabilities in every major operating system and browser. The public version may be called Claude Fable. Per Alex Heath at Sources.

Table of Contents

What Is Mythos

Claude Mythos is Anthropic’s most advanced model. Originally restricted to 200 organizations through Project Glasswing (a cybersecurity consortium), Mythos is not just a better chatbot — it is a fundamentally different class of AI capability:

Claude Mythos — Key Capabilities

10,000

Critical vulnerabilities
found in first month

Every

Major OS + browser
zero-day exploited

200

Organizations with
access (Project Glasswing)

15+

Countries in
the consortium

Partners include: NATO, Samsung, SK Hynix, Okta, ENISA (EU cybersecurity agency)

Claude Fable: The Public Version

Per Alex Heath at Sources (reporting from WWDC), Anthropic is about to release the public version of Mythos — potentially named Claude Fable. This would make Mythos-class capabilities available to all Claude customers, not just the Glasswing consortium.

Anthropic delayed the public release due to security risks — the model’s ability to autonomously develop professional-grade cyberattacks raised questions about whether public access was responsible. The company says it has made “swift progress on developing safeguards.”

The Map of AI Read

In the Map of AI, this is a Layer 4 (Models) event with Layer 6 (Enterprise Applications) implications:

Layer 4 — Models: Mythos represents a capability jump that exceeds Opus 4.8 in reasoning, coding, and autonomous operation. If released publicly, it shifts the frontier of what commercial AI models can do.
Layer 6 — Enterprise: The Glasswing consortium (NATO, Samsung, Okta, ENISA) signals that the first killer app for frontier AI may not be chatbots — it may be cybersecurity. Finding 10,000 critical vulnerabilities in a month is not an incremental improvement. It is a structural shift in how software security works.

The Product Overhang

Anthropic has been sitting on Mythos since April. 200 organizations have access. The public does not. This is a textbook Product Overhang:

Capability exists (10,000 zero-days, professional-grade autonomous exploitation)
Distribution is restricted (200 orgs through Glasswing)
Safeguards are the valve (Anthropic delayed release until guardrails were ready)

When Fable launches publicly, the overhang releases. Every developer, security team, and enterprise customer gets access to capabilities that were military-grade two months ago.

The Timing

Consider what happened in the last 48 hours:

Apple WWDC: Siri rebuilt on Gemini — harness model validated

OpenAI S-1: $852B valuation, $2B/month revenue — Layer 4 goes public

Goldman Sachs: $7.6T AI capex projected — infrastructure confirmed

AMD: £2B UK AI investment — Layer 2 fracturing

Anthropic Mythos/Fable: Public release imminent — the frontier just moved again

The AI economy is not gradually shifting. It is snapping — overhang after overhang releasing in rapid succession. Five structural events in 48 hours. Every layer of the Map of AI is moving simultaneously.

Frameworks:
The Map of AI — 9 Layers of the AI Economy
Product Overhang Doctrine
Harness Theory

Sources: Alex Heath / Sources newsletter, CNBC, Bloomberg, TechCrunch, The Register, BleepingCompute — as explored in the economics of AI compute infrastructure — r, Anthropic red.anthropic.com