Anthropic announced on Tuesday the launch of a collaborative initiative with several major technology companies, allowing partners to preview an advanced AI model with cybersecurity capabilities.
According to information released by Anthropic, the program, named "Project Glasswing," will grant select organizations access to its unreleased general-purpose AI model "Claude Mythos Preview" for defensive cybersecurity work. Partners include technology and cybersecurity firms such as Amazon, Microsoft, Apple, Alphabet, NVIDIA, CrowdStrike, and Palo Alto Networks.
The model's capabilities have drawn significant attention. In a blog post, Anthropic stated that the Mythos Preview has already identified "thousands" of significant vulnerabilities in operating systems, web browsers, and other software. The company reported that the security flaws discovered by the model include a 27-year-old bug in OpenBSD and a 16-year-old vulnerability in FFmpeg that remained undetected even after automated testing tools were run five million times.
Notably, Mythos was not specifically trained for cybersecurity tasks, yet its "powerful agent programming and reasoning capabilities" enable it to excel in the cybersecurity domain. Anthropic indicated that the model can identify vulnerabilities and develop related exploit code "completely autonomously," requiring no human guidance.
A phased release strategy has been adopted due to concerns about potential misuse of the model. The company stated it currently has no plans for a full public release and will only provide access to project partners and approximately 40 additional organizations responsible for critical software infrastructure.
Anthropic has committed up to $100 million in usage credits for the project and will donate $4 million to open-source security organizations. Participating organizations are required to share their findings with the broader industry.
The announcement comes as the cybersecurity industry is highly focused on the rise of AI-driven cyberattacks. A study conducted by IBM and Palo Alto Networks revealed that 67% of 1,000 surveyed executives reported being targeted by an AI-based attack within the past year.
Anthropic stated it has been engaged in "ongoing discussions" with U.S. government officials regarding the capabilities of the Mythos Preview. Previously, the company was involved in legal disputes with the Trump administration after refusing to authorize the U.S. Department of Defense to conduct autonomous targeting or surveillance of citizens.
Comments