Claude Mythos Leak: Model 'Too Dangerous for Public' Falls Into Unauthorized Hands

Claude Mythos Leak: Model "Too Dangerous for the Public" Reached Unauthorized Users

Anthropic failed to keep access to the Claude Mythos model under control: according to Bloomberg, published on April 21, 2026, a "small group of unauthorized users" gained access to the system, which the company itself had been calling too dangerous for public release for weeks.

How the Model Reached the Wrong Hands

Unauthorized access to Mythos, according to Bloomberg, persisted from the day Anthropic first announced plans to open the model for testing to a limited circle of companies as part of the Project Glasswing initiative, dedicated to cybersecurity. Even earlier, the very existence of Mythos became known from a leak—that is, this is already the second story in a short time when information about the model seeps beyond the company against its will.

• Mythos existence leak occurred before the official model announcement
• Anthropic announced Project Glasswing—a plan to test Mythos with a limited circle of partners
• According to Bloomberg on April 21, 2026, a "small group of unauthorized users" gained access
• Anthropic confirmed it is investigating the incident

Why This is Awkward for Anthropic

Anthropic spent weeks insisting that Claude Mythos was so powerful in cybersecurity tasks that public release of the model would simply be dangerous—and at the same time was preparing closed test access for select partners. The Verge calls what happened humiliating for a company that has built its brand on taking AI safety seriously and constantly emphasizing cybersecurity capabilities of its models. Anthropic said it is investigating the incident, but did not disclose either the scale of the leak or exactly how outsiders gained access.

"A small group of unauthorized users" gained access to

Mythos, according to Bloomberg on April 21, 2026.

What is Known About Project Glasswing

Project Glasswing is a program announced by Anthropic, under which the company planned to provide a limited circle of companies access to Claude Mythos for testing the model's cybersecurity capabilities. The very existence of Mythos became publicly known earlier than intended—from a leak rather than from an official Anthropic announcement, which already cast a shadow on the release preparation process.

What This Means

The Mythos story shows: the stricter a company controls access to a model and the louder it proclaims its danger, the more painful any leak looks—Anthropic found itself in a situation where actual access practices diverged from public rhetoric about safety. A company that promised to hand out Mythos only to select partners under control now has to explain how people without permission ended up with access.