It’s a high-profile announcement making waves in the cybersecurity world. Anthropic has just unveiled Claude Mythos.
Dubbed Mythos Preview, this frontier model boasts capabilities in detecting software vulnerabilities that surpass “all humans except the most seasoned experts.”
To govern its use and steer it toward defense, Anthropic simultaneously launched the Project Glasswing; a coalition bringing together AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks.
Mythos Preview is described as a frontier model
The results presented are dizzying. In a matter of weeks, Mythos identified thousands of zero-day vulnerabilities across every major operating system and major web browser.
Among the documented cases:
- A 27-year-old vulnerability in OpenBSD, a system considered among the most secure in the world, that could remotely crash any connected machine;
- A 16-year-old vulnerability in FFmpeg, in a line of code that automated tools had executed five million times without ever detecting it;
- Several chained flaws in the Linux kernel enabling an attacker to take full control of a machine from a mere user account.
Un outil offensif mis au service de la défense
The tension sits at the heart of the project: a model this powerful could be devastating in the wrong hands. That is precisely why Project Glasswing was designed: to give defenders a head start.
Founding partners will use Mythos Preview as part of their defensive security work. On its side, Anthropic commits to sharing the lessons learned so that the entire industry can benefit.
More than 40 additional organizations that build or maintain critical software infrastructure have also been granted access to the model, to analyze and secure their proprietary and open-source systems.
For those who wish to access the model after the preview phase, Mythos Preview will be available at $25 per million input tokens and $125 per million output tokens, accessible via the Claude API, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry.
$100 Million in Credits and Open-Source Donations
Anthropic has committed $100 million in usage credits for Mythos as part of Project Glasswing, as well as $4 million in direct donations to open-source security organizations. Specifically, $2.5 million has been disbursed to Alpha-Omega and the OpenSSF via the Linux Foundation, and $1.5 million to the Apache Software Foundation.
Jim Zemlin, CEO of the Linux Foundation, underscored the challenge facing open-source maintainers, often ill-equipped to handle security issues. By giving them access to AI models capable of identifying and patching vulnerabilities at scale, the Glasswing Project offers a credible path for AI-enhanced security to become an accessible ally for all maintainers, not just those with large security teams.
Mythos Preview also posts the best scores ever recorded on several coding and reasoning benchmarks: 93.9% on SWE-bench Verified, 77.8% on SWE-bench Pro, and 94.6% on GPQA Diamond.
No general availability planned for now
Anthropic does not plan to make Mythos Preview accessible to the general public. The long-term goal is to enable large-scale deployment of models of this caliber, but that first requires building sufficiently robust guardrails.
The scale-up plans to roll out these new protections with the upcoming Claude Opus model, which will allow them to improve and refine safeguards on a model that presents a lower risk than Mythos Preview.
