How the built-in browsers in five AI coding tools broke basic browser rules (theainewsreport.com)

A recent report by Forever Security titled EMBRACE has revealed critical vulnerabilities in the built-in browsers of five prominent AI coding tools: Cursor, GitHub Copilot, OpenAI Codex, Visual Studio Code (VS Code), and Claude Desktop. These browsers were found to violate fundamental web security principles, including the ability to run unauthorized commands, read local files outside of their project directories, and spoof address bars. While Cursor and GitHub Copilot have patched their vulnerabilities, Codex, VS Code, and Claude Desktop remain unpatched as of October 29, 2026. The significance of these findings lies in the potential security risks they pose to users, which could lead to unauthorized access to sensitive information, such as SSH keys and environmental variables. The tests revealed alarming capabilities, such as Cursor's ability to execute commands with a single keystroke and Codex's ability to read files across the disk from local HTML previews. Users of these tools are advised to restrict the use of untrusted projects and run searches for specific exploit indicators within their code bases to mitigate risks before fixes are implemented. The research underscores the pressing need for enhanced security measures in AI-driven development environments.