www.it-connect.fr
Anthropic Says Claude Mythos Will Reach the Public in Coming Weeks
Since April 2026, the Claude Mythos Preview model developed by Anthropic has been reserved for a small circle of organizations and security researchers because of the risks it may pose to the software landscape. Nevertheless, Anthropic has decided to roll it out to the general public in the coming weeks.
In a new blog post, Anthropic unveiled the new version of Claude Opus, namely Claude Opus 4.8. This was expected, since Anthropic aims to release a new version at a regular pace, every two months to be exact. It therefore succeeds Claude Opus 4.7, is intended to be more powerful than previous versions (and the competition), and also more honest with its user. But above all, Anthropic also dropped a small bombshell in this article: Claude Mythos will soon be available to the general public.
The Arrival of Claude Mythos
When it was announced last April as part of the Glasswing project, Anthropic revealed Claude Mythos Preview, a model above all the others and, above all, one that was not meant for everyone. In fact, Anthropic chose to restrict access to a very selective group of partners: Google, VMware, Mozilla, Palo Alto Networks, and others...
The reason: Claude Mythos represents major cybersecurity risks, notably because of its ability to identify vulnerabilities in any application. It has already proven itself in partners' products, with an explosion in the number of flaws discovered in Mozilla Firefox, Google Chrome, and more.
So, the day Claude Mythos becomes accessible to the general public, it will benefit developers, security teams, but also cybercriminals. That day will arrive in the coming weeks.
"Models at this level of performance require strengthened cybersecurity measures before they can be made available to the general public. We are making rapid progress in developing these safety measures and hope to offer models from the Mythos category to all our customers in the weeks ahead," Anthropic explains in its article.
Anthropic is therefore putting safeguards in place to prevent things from going completely off the rails the day this model becomes available to everyone. The goal is to limit misuse, but that should not prevent the discovery of vulnerabilities: it seems particularly comfortable in that area.
Claude Mythos is considered by Anthropic to be "a new category of AI models with intelligence even greater than Opus." - The improvements are at two levels:
- A major improvement in reasoning applied to computer code.
- A much higher level of autonomy than the current flagship model.
Who Will Benefit, and When?
There are still a few unknowns. There is no exact date for Claude Mythos availability. We also do not know who will actually be able to use it: will it be reserved for those with a Claude Max subscription? That is not impossible.
We also do not know whether the final public version will really be the same as the one currently provided to cybersecurity researchers (apart from the new safeguards introduced).
What is certain is that Claude Mythos is probably the model generating the most interest right now.
