Anthropic showed off Project Glasswing, an AI-based cybersecurity project that uses Claude Mythos This article explores project glasswing ai. . The goal of the project is to find and fix security holes in important software.
Anthropic has set aside up to $100 million in usage credits for Mythos Preview and given an extra $4 million directly to open-source security groups. Anthropic said that the model has found thousands of high-severity zero-day vulnerabilities in all major operating systems and web browsers.Adversa, a security company, said that Anthropic's model is the most powerful AI model ever made. The company said that Project GlassWing is a "critical endeavor" to use cutting-edge models for defense purposes before they are used by hostile groups.
It also made it clear that Anthropic did not specifically teach its model these skills; they just happened as a result of improvements in code, reasoning, and autonomy in general. The company also talked about another security breach that let out almost 2,000 source code files and more than half a million lines of code related to Claude Code for about three hours. The leak revealed a serious flaw that lets commands with more than fifty subcommands get around security measures.
Anthropic quickly fixed this bug in the most recent version of Claude Code, version 2.1.90, which came out not long ago. If a developer sets "never run rm," the command will be blocked when run by itself, but it will run without any problems if it is preceded by fifty harmless statements.
In this case, the security policy basically disappears. The Anthropic team had a problem with performance when they tried to check every subcommand. This caused the interface to freeze and used up resources.
To fix this, they decided to stop checks after 50 iterations. The company said, "This trade-off put speed ahead of security, and safety was sacrificed for cost." The draft said that it was the most powerful AI modelever ever made.
