Anthropic appears to be preparing for a public release of “Mythos,” which was announced in April as a limited model that poses significant security risks to private and public software.
On April 7, Anthropic announced the Mythos in a preview and called it a new frontier model with dramatically improved capabilities in computer security operations.
Anthropic said the Mythos model shows significant improvements in coding and autonomy, over its current model, Opus 4.7.
Code optimization is nothing new for AI models, but in the case of Mythos, Anthropic found that the model can automatically optimize an effective cyberattack at a very high level.
The company also said that its release poses a serious risk to the global digital infrastructure.
“The benefit will be to the party that can benefit the most from these tools,” warns Anthropic.
“In the short term, this could be the attackers, if the border labs are not careful how they release these models. In the long term, we expect that it will be the defenders who will direct resources more effectively and use these models to fix bugs before the new code ever goes live.”
In order to prevent attackers from using a large volume of unblocked vulnerabilities in popular applications, such as Firefox, Anthropic decided against the public release of the Mythos model until it prepares a powerful monitoring system.
It looks like Anthropic may have developed a stronger surveillance system, as Claude Code and Claude Security now have references to the Mythos model.
In fact, some users briefly saw the switch to enable Mythos in the public version of the Claude Code before it was removed from the Internet.
The model is called claude-mythos-1-preview, and it appeared briefly in the public version of Claude Security.
This confirms that Anthropic is preparing the model for public release, but it is not clear if it will be available in all subscription categories.
Anthropic says it’s working with companies on potential AI-powered inventions
Anthropic has confirmed that it is working on a new project called “Glasswing,” in which the AI startup collaborates with other companies to protect the world’s most important software from potential AI-driven incidents.
This program uses the unreleased Claude Mythos Preview, and has been able to support up to 50 organization partners.
The Mythos model was able to uncover 10,000 high or critical vulnerabilities in its first month alone, which explains why Anthropic held its release public.
Anthorpic currently offers Claude Opus 4.7, Opus 4.6, Opus 4.5, Sonnet 4.6, and Haiku 5.5.
Automated testing tools deliver real value, but they’re designed to answer one question: can an attacker deploy on a network? They are not designed to check that your controls are blocking threats, your firewall detection, or your cloud configs.
This guide covers the 6 areas you really need to verify.
Download Now
