vscode-java CVE-2026-12856

Exploitation requires the victim to open an attacker-supplied Java file/project in vscode-java, display a JavaDoc hover popup for an attacker-controlled symbol, and click a maliciously crafted link embedded in that hover's Markdown; the workspace must be in trusted (non-Restricted) mode for the injected VS Code commands to execute. Limiting factors: mandatory user interaction (the click, reflected as UI:R), the need to first lure the victim into opening the malicious project, and VS Code Workspace Trust - projects opened in Restricted Mode block the command execution. No standing authentication to any service is required (PR:N), so any developer who processes untrusted Java code is in scope.

The CVSS 3.1 vector (CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H, base 8.8) indicates a network-reachable, low-complexity, unauthenticated path with high confidentiality, integrity and availability impact, gated only by required user interaction. The dominant real-world limiter is UI:R combined with the description's 'trusted workspaces' qualifier: the victim must open the malicious project, hover JavaDoc, and click the embedded link, and the workspace must be in trusted mode for commands to run. No EPSS score was provided in the input, and there is no KEV listing and no POC reference, so this is currently a credible-but-not-demonstrated risk rather than confirmed in-the-wild exploitation. Note one consistency caveat: AV:N reflects remote delivery of the malicious file, but actual code execution is local to the developer's machine after they open the project and click - defenders should treat this as a social-engineering / malicious-repository threat, not a remote drive-by.

An attacker publishes a seemingly legitimate Java library or sample project to a public repository, with a class whose JavaDoc comment embeds a malicious command: link disguised as ordinary documentation. A developer clones and opens it in VS Code with the Java extension in a trusted workspace, hovers a method to read its docs, and clicks the link - executing arbitrary VS Code commands that can run code and compromise the host. No POC is referenced in the available data, and the low attack complexity (AC:L) means only basic social engineering to get the click is required.