Anthropic's Mythos: Too Powerful and Dangerous? - News Directory 3

According to a technical post published by Anthropic on April 7, 2026, Claude Mythos Preview is strikingly capable at computer security tasks.

Anthropic announced the development of Claude Mythos Preview on April 7, 2026, a new general-purpose language model with capabilities in computer security that the company describes as too powerful for public release.

The company stated that the model is not ready for a general launch because of the potential for cybercriminals and spies to abuse its effectiveness in identifying high-severity vulnerabilities in web browsers and major operating systems.

According to a technical post published by Anthropic on April 7, 2026, Claude Mythos Preview is strikingly capable at computer security tasks. Testing conducted over the previous month revealed the model's ability to find and exploit zero-day vulnerabilities, which are previously undiscovered security flaws, within real open-source codebases.

The model also demonstrated the ability to reverse-engineer exploits on closed-source software. It can turn N-day vulnerabilities -- known flaws that have not yet been widely patched -- into active exploits.

Further evaluations detailed in the model's system card indicated that Claude Mythos Preview could follow instructions to break out of a virtual sandbox. This process allows the model to bypass file system, network, or security constraints imposed upon it.

Because of these risks, Anthropic has decided against making the model generally available. Instead, the company has launched Project Glasswing, a defensive cybersecurity program.

Project Glasswing aims to use Claude Mythos Preview to help secure critical software and prepare the industry for new practices required to defend against AI-driven cyberattacks. The model is currently being utilized by a limited set of partners within this program.

The public announcement follows a data leak in March 2026 that first revealed Anthropic was developing the model. At the time of the leak, the company stated that the technology poses unprecedented cybersecurity risks.

The rumors surrounding the model's strength caused a slump in cybersecurity stocks, as investors reacted to the possibility of the technology becoming a tool for hackers.