Anthropic finally offering its Mythos AI model to the public
What's the story
Anthropic has launched two new artificial intelligence (AI) models, Claude Fable 5 and Claude Mythos 5. The company claims these models are more capable than the previous Mythos Preview model released in April to a select group of tech industry partners. The limited release was due to fears that the model's capabilities could be misused by malicious actors to create hacking tools that could surprise institutions.
Restricted release
Limited availability of Claude Mythos 5
Currently, Anthropic is only making Claude Mythos 5 available to a select group of industry partners, many of whom had access to the earlier Mythos Preview. The company is also working with the US government on this rollout. This move highlights Anthropic's commitment to ensuring that its advanced AI technology is used responsibly and securely by trusted partners in the tech industry.
Public launch
Claude Fable 5 has 'guardrails'
Unlike its counterpart, Claude Fable 5 has been publicly released. It comes with "guardrails" that will prevent it from answering many user queries related to cybersecurity, biology, and chemistry at launch. These queries will be redirected to an older AI model, Claude Opus 4.8. If Anthropic suspects a user is trying to conduct distillation on Claude Fable 5 using this method of training a smaller AI model off a larger one's responses, those requests will also be redirected to Opus.
Cautious strategy
Why are there guardrails?
Diane Penn, Anthropic's head of product management, told WIRED that the company has been grappling with how to handle Mythos's software vulnerability-discovery abilities, and other advanced capabilities since before its April release. She said testing and user input since then helped refine their strategy. The protective mechanism is designed to be overly cautious for now, meaning some user queries may be redirected to the less capable AI model even if they're benign.
Expansion strategy
What about availability?
Along with providing Claude Mythos 5 to Project Glasswing partners, Anthropic is also giving access to "select biology researchers." The company said it is providing unrestricted versions to these small groups of customers "until our trusted access program is available," indicating plans for further expansion. Since the Mythos launch in April, Anthropic has repeatedly emphasized that its competitors will eventually offer models with Mythos-level capabilities.
Cybersecurity concerns
Cybersecurity risks associated with AI models
The ability of Claude Mythos and other new AI models to design hacking tools that can find and exploit vulnerabilities in both new and legacy software has raised cybersecurity concerns. Tech companies and governments around the world are now rushing to secure their software defenses before such advanced AI models are made widely available to potential attackers.