Anthropic shared valuable information about the performance of the Claude Mythos Preview model developed within the scope of the Glasswing project. The company announced that this new model detected more than 10 thousand critical security vulnerabilities in just one month.

This study, which aims to reduce the risks of cyber attacks, aims to protect the world’s most critical software. In the process carried out with approximately 50 different partners, high and critical vulnerabilities in software are targeted.

Security Scans with Claude Mythos

According to the statement made on the official blog page, the Claude Mythos Preview model plays an important role in the field of cyber security. The company states that with the help of this model, they managed to find many security vulnerabilities in critical software.

Business partners such as Cloudflare are also actively involved in this process. While Cloudflare detected a total of 2,000 flaws using Claude Mythos Preview, it is stated that 400 of them were at high or critical value.

Mozilla is also among the institutions involved in the project. Mozilla announced that it found 271 different vulnerabilities in its investigations on the Firefox browser with the help of Claude Mythos Preview.

Massive Scan of Open Source Projects

Anthropic carries out extensive work not only with corporate partners but also on open source projects. The company used the Claude Mythos Preview version to scan 1,000 different open source projects over the past few months.

As a result of this large-scale scanning, a total of 23,019 vulnerabilities were recorded. The company confirmed that 6,202 of these vulnerabilities were high and critical vulnerabilities.

The Glasswing project was launched with the aim of preventing the use of artificial intelligence models in cyber attacks. This technology is positioned as a critical tool to make software more reliable.

How do you think these types of models will change security standards in the software world in the future?