CVE-2026-70335

Copilot could run commands from untrusted content without confirmation

Microsoft · Visual Studio Code

First documented disclosure · Review checkpoint

Summary

Instructions in a web page, repository file or tool response could make the agent execute local commands.

Severity and exploitation evidence

High · 7.8/10 · CVSS 3.1. Severity source.

Exploitation not established: Microsoft marks Exploited: No. The reviewed material does not establish malicious exploitation against users.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

Untrusted content becomes an operating-system command without the intended confirmation.

Conditions required

The user runs the AI agent against attacker-supplied content; no attacker account is required.

Potential impact

Code execution as the signed-in user.

Affected versions and configurations

Visual Studio Code: 1.0.0 to before 1.132.1.

Fix and mitigation status

Visual Studio Code: 1.132.1.

Practical next steps

Install the vendor-listed corrected build for the applicable product and platform.

Operating lesson

Enforce the affected trust boundary independently of AI-generated content, caller identity and user-interface safeguards.

Evidence and limits

Historical backfill from a November 19–20 CVE update batch. Earliest reviewed public date is 2026-08-11; CNA initial publication is 2026-08-11, and current CNA datePublic is 2026-08-11T14: 00: 00.000Z. Microsoft CVSS 3.1 base score is 7.8; separate MSRC severity label: Important. The manufacturer marks Exploited: No in this reviewed entry. That is an attribution, not an assurance against future exploitation.

Sources