Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Anthropic released two new AI models called Claude Fable 5 and Claude Mythos 5 on Tuesday, which the company says have greater capabilities than… Mythology Preview this model Released in April To a limited group of technology industry partners. Anthropic He said the limited initial release stems from concerns that the model’s capabilities could be exploited by bad actors to develop hacking tools that could detect… Defenders Suddenly.
Anthropic is currently only releasing Claude Mythos 5 to a limited group of industry partners, many of whom have gained access to the Mythos Preview, and the company says it is cooperating with the US government in the rollout.
Claude Fable 5, which has been released publicly, uses the same basic model as Mythos 5, but will have “guardrails” at launch, the company said Tuesday, which will prevent the model from answering many user questions related to cybersecurity, biology and chemistry. These requests will instead be redirected to the older AI model, Claude Opus 4.8. If Anthropic suspects that the user is attempting to do so distillation– Training a smaller AI model on the responses of a larger AI model – In Claude Fable 5, these requests will also be forwarded to Claude Opus 4.8, the company says.
In an interview with WIRED, Anthropic’s head of product management, Diane Penn, said the company had been grappling with the question of how to approach Mythos’ software vulnerability detection and other advanced capabilities since before its release in April, but that testing and user input since then has helped refine the strategy.
“We try to make improvements in a meaningful way, even if we don’t have the perfect[solution]to start with every use case,” Ben says. “Of all the different approaches, this emerged as the most viable and best. We ended up feeling like this was the best product option for users to get the most value out of Fable 5.”
For now, Ben says the protection mechanism is built to err on the side of caution, meaning some user queries may be routed to the less capable AI model even if they are benign. Over time, Anthropic hopes to make its classification tools more accurate, but Penn says this was the only safe way the company could release the model on a large scale at this time.
In addition to offering Claude Mythos 5 to Glasswing project partners, it is also offering access to “selected biology researchers,” the company said Tuesday. In addition, Anthropic pointed out in Blog post About Tuesday’s launch it is offering unrestricted versions to these small groups of customers “until our trusted access program is available,” hinting at future plans to expand access further. Since the launch of the Mythos in April, Anthropic has repeatedly confirmed that its top-tier competitors in both the privates and even openweight spaces will inevitably offer models with Mythos-level capabilities.
The ability of Cloud Mythos and other new AI models to design hacking tools that can find and exploit vulnerabilities in both new and old software has forced technology companies and governments around the world to secure their software defenses before AI models of this level become widely available to attackers. Anthropic first launched Mythos to industry partners under a consortium called Project Glasswing, with the idea that this could give members an early start in preparing their own systems and weighing global solutions to the threat before a wider release.
Anthropy wrote in to update About the Glasswing project last week: “We are working as quickly as possible to safely release Mythos-level capabilities into public access. To do this, we will need very strong safeguards that prevent abuse of the model’s cyber capabilities — safeguards that we (and, as far as we know, all other AI developers) have not yet developed.”
