What Happened
On April 7, 2026, Anthropic confirmed the existence of Claude Mythos — calling it "by far the most powerful AI model we have ever developed" and a "step change in capabilities." The model was first revealed through a data leak when a CMS misconfiguration exposed draft blog posts.
But here's the twist: they're not releasing it.
NEWS: Anthropic's new model, Claude Mythos, is so powerful that it is not releasing it to the public. Instead, it is starting a 40-company coalition, Project Glasswing, to allow cybersecurity defenders a head start in locking down critical software.
— Kevin Roose (@kevinroose) April 7, 2026
Project Glasswing
Instead of a public release, Anthropic launched Project Glasswing — a 40-company coalition including Amazon, Apple, Microsoft, CrowdStrike, Cisco, and Palo Alto Networks. The goal: use Mythos to find and fix vulnerabilities in critical software before attackers can exploit them.
We do not plan to make Mythos Preview generally available. Our goal is to deploy Mythos-class models safely at scale, but first we need safeguards that reliably block their most dangerous outputs.
— Anthropic (@AnthropicAI) April 7, 2026
The Capabilities
The numbers are staggering:
- 93.9% on SWE-bench Verified (vs Opus 4.6 at 80.8%)
- 97.6% on USAMO math olympiad (vs 42.3%)
- 181 successful Firefox exploits (vs 2 for Opus 4.6)
- 100% solve rate on Cybench CTF challenges
Over the past few weeks, Mythos Preview identified thousands of zero-day vulnerabilities — previously unknown to the software developers — in every major operating system and every major web browser.
What This Means for AI Orchestrators
If a model can autonomously find thousands of zero-day vulnerabilities, every AI system you build needs to be hardened from day one. This isn't theoretical — it's happening now.
AI security is no longer optional. Input validation, permission boundaries, output verification, and monitoring are the orchestrator's non-negotiables.