AI-driven integrated development environments (IDEs) have completely changed the software development landscape.

From basic autocompletion engines, tools like GitHub Copilot, Gemini CLI, and Claude Code have developed into autonomous agents that can carry out tasks.

However, by integrating these agents straight into traditional IDE systems that were not initially intended for such autonomy, companies unintentionally increase attack surfaces, creating a security vulnerability because of this quick drive for productivity.

The recently identified vulnerability class, known as “IDEsaster,” takes advantage of how AI agents interact with the fundamental components of standard IDEs like Visual Studio Code and JetBrains.

This attack chain uses IDE features, like configuration files and workspace settings, to carry out malicious actions, in contrast to traditional vulnerabilities that target a specific tool’s weakness.

Attackers might circumvent common security barriers by altering these fundamental components, converting useful features into conduits for data exfiltration and remote code execution.

Following a thorough investigation into the security posture of AI coding helpers, MaccariTA security analysts discovered this concerning tendency. All the evaluated applications were susceptible to this new type of attacks, according to their analysis.

With over 30 distinct vulnerabilities discovered and 24 CVEs given across market-leading products, the repercussions are enormous.

Manipulating IDE configuration files to accomplish Remote Code Execution (RCE) is the most extreme kind of IDEsaster. In this case, an attacker uses prompt injection to fool the AI agent into changing important settings files, like.idea/workspace.xml in JetBrains IDEs or.vscode/settings.json in Visual Studio Code.

Unlike earlier vulnerabilities, this one targets global IDE settings instead of simply agent-specific setups. For example, with Visual Studio Code, an attacker can instruct the agent to install malicious code into a seemingly innocuous file, such as a Git hook sample.