Anthropic Restricted Access to AI Vulnerability Model Mythos to 200 Partners

Anthropic PBC stated that its AI model Mythos is so effective at finding vulnerabilities in software and computer systems that the company decided against a public release: as of early June 2026, only 200 partner organizations have access to Mythos.

What Mythos can do

The model specializes in finding vulnerabilities in software and computer systems—a task that the cybersecurity industry calls security research or penetration testing. According to Anthropic, Mythos performs this task so effectively that the company considered a mass public release of the tool too risky a step.

• Developer of the model — Anthropic PBC
• Mythos specialization — finding vulnerabilities in software and computer systems
• As of early June 2026, 200 partner organizations have access to the model
• The company does not plan a public release for a broad audience

Why the company restricted access

Tools that find vulnerabilities are by nature dual-use: the same discoveries that help defenders patch holes in systems turn into a ready-made attack plan in the wrong hands. Anthropic explains the access restriction precisely by this risk—according to the company's statement, if a tool of such power falls into the wrong hands, it could facilitate data theft or disruption of critical infrastructure.

Instead of an open launch, Anthropic chose a controlled access model: Mythos is tested only by vetted partner organizations, not by a broad audience of developers or researchers.

What this means

The Mythos story is an instructive example of a dilemma that AI laboratories increasingly face when developing cybersecurity tools: the more effective the model is at finding vulnerabilities for defenders, the more dangerous it is in the hands of attackers. Anthropic chose a cautious scenario—limited distribution instead of the risk that a tool of this level would amplify cyberattacks on critical infrastructure.

Frequently asked questions

How many organizations have access to Mythos?

As of early June 2026, Anthropic provided access to Mythos to only 200 partner organizations; the model is not available to the general public.

Why did Anthropic not release Mythos to open access?

The company fears that such an effective vulnerability-finding tool in the wrong hands would help malicious actors steal data or disrupt critical infrastructure.