Reports indicate that access to 'Claude Mythos Preview,' which is only available to a select group of organizations, has fallen into the hands of unauthorized users.

Anthropic's AI model, ' Claude Mythos Preview ,' announced on April 7, 2026, possesses extremely high cyberattack capabilities. To advance countermeasures against AI-based cyberattacks, it has been withheld from public release and is only available to a limited number of organizations. Bloomberg has reported that an unauthorized user has obtained access to this Claude Mythos Preview.

Anthropic's Mythos AI Model Is Being Accessed by Unauthorized Users - Bloomberg
https://www.bloomberg.com/news/articles/2026-04-21/anthropic-s-mythos-model-is-being-accessed-by-unauthorized-users

Unauthorized group has gained access to Anthropic's cyber exclusive tool Mythos, report claims | TechCrunch
https://techcrunch.com/2026/04/21/unauthorized-group-has-gained-access-to-anthropics-exclusive-cyber-tool-mythos-report-claims/

Anthropic's Mythos model accessed by unauthorized users, Bloomberg News reports | Reuters
https://www.reuters.com/technology/anthropics-mythos-model-accessed-by-unauthorized-users-bloomberg-news-reports-2026-04-21/

Claude Mythos Preview is an AI model that excels at discovering and exploiting vulnerabilities. Tests by Anthropic have shown that it can autonomously discover vulnerabilities in OpenBSD and Linux and create attack exploits. Analysis by the AI Security Institute, a UK government agency, has also confirmed that Claude Mythos Preview possesses autonomous attack capabilities.

Tests by a UK government agency revealed that 'Claude Mythos Preview' can autonomously execute a complete network takeover attack - GIGAZINE

Anthropic recognizes the risks of Claude Mythos Preview and has refrained from making it publicly available. Instead, it has made it available to a limited number of organizations to help them defend against attacks using AI with comparable capabilities.

However, according to Bloomberg, a small number of users have illegally obtained access to the Claude Mythos Preview. The person who provided the information to Bloomberg reportedly presented screenshots and live demos as evidence.

The fraudulent user was a member of a private Discord server dedicated to searching for information on unreleased AI models, and used bots to analyze information posted by Anthropic on public services such as GitHub. Furthermore, it is believed that they obtained access to Claude Mythos Preview by combining methods such as 'abusing access rights granted while working as a subcontractor for Anthropic' and 'running common internet research tools used by cybersecurity researchers.'

Anthropic told Bloomberg, 'We are investigating reports of unauthorized access to Claude Mythos Preview through third-party vendor environments.'

Furthermore, the high attack capabilities of Claude Mythos Preview pose a threat to open-source projects, and some projects are now migrating parts of their code to closed source.

Open-source software developers decide to move to closed-source software due to security risks associated with AI - GIGAZINE

On the other hand, Mozilla has reported using Claude Mythos Preview to discover and fix 271 vulnerabilities hidden in Firefox, confirming that it is a useful tool not only for attackers but also for defenders.

Anthropic's Mythos discovers 271 security vulnerabilities in Firefox 150; Mozilla calls it 'good news for defenders' - GIGAZINE