Anthropic unveils Claude Mythos for vulnerability detection but restricts access
Anthropic has launched a new AI model, Claude Mythos, designed to detect software vulnerabilities. However, the company has chosen not to offer public access, limiting the model’s availability to major partners such as Amazon, Google, Microsoft, and Nvidia.
Source Anthropic
According to Anthropic, Claude Mythos significantly outperforms humans in finding bugs and has already uncovered critical vulnerabilities, including ones that remained undetected for 27 years. This highlights the model’s strong potential in the field of cybersecurity.
The comparative benchmark tables show Claude Mythos achieving notably higher scores across various tests, including SWE-bench Verified (80.6%) and SWE-bench Pro (30.5%) compared to other state-of-the-art models like GPT-3.5-turbo and Gemini 3.1 Pro. This demonstrates improved capabilities in handling multilingual and multimodal data.
Anthropic is an innovative AI company focused on developing secure and reliable AI models. It has garnered attention from industry leaders by creating unique solutions that address pressing cybersecurity challenges.
In summary, the launch of Claude Mythos marks a significant advancement in automated security testing tools, although restricted access currently limits its widespread adoption. The company is expected to broaden user access over time and make its innovations more widely available.
Looking ahead, models like Claude Mythos could transform cybersecurity practices by enhancing the speed and accuracy of vulnerability detection, thereby reducing the risks associated with cyberattacks.
