CVE-2026-95182
Received Received - Intake

Untrusted Search Path in DeepSeek Harness

Vulnerability report for CVE-2026-95182, including description, CVSS score, EPSS score, affected products, exploitability, helpful resources, and attack-flow context.

Publication date: 2026-10-16

Last updated on: 2026-10-16

Assigner: MITRE

Description

An untrusted search path in DeepSeek Harness (dsh) commit c291e796 allows attackers to execute arbitrary code via supplying and opening crafted executable files in place of legitimate content.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published
2026-10-16
Last Modified
2026-10-16
Generated
2026-10-16
AI Q&A
2026-10-16
EPSS Evaluated
N/A
NVD
EUVD

Affected Vendors & Products

Showing 1 associated CPE
Vendor Product Version / Range
n/a n/a n/a

Helpful Resources

Exploitability

CWE
CWE Icon
KEV
KEV Icon
CWE ID Description
CWE-426 The product searches for critical resources using an externally-supplied search path that can point to resources that are not under the product's direct control.

Attack-Flow Graph

AI Quick Actions

Instant insights powered by AI
Executive Summary

CVE-2026-95182 is an Untrusted Search Path vulnerability in DeepSeek Harness (dsh) developer preview. It allows attackers to execute arbitrary code by replacing legitimate executables with malicious ones in writable directories like node_modules/.bin. The issue occurs because dsh spawns MCP servers using unqualified command names, inheriting the host's PATH environment variable. An attacker can plant a malicious executable with the same name as a legitimate command, which dsh will execute instead when spawning or reconnecting to MCP servers.

Detection Guidance

Check for unexpected executables in writable directories like node_modules/.bin. Inspect PATH environment variables for suspicious entries. Monitor MCP server spawns for unusual command executions or reconnect retries.

Impact Analysis

This vulnerability enables persistent, infrastructure-level compromise. Attackers can gain long-lived access to MCP servers with the operator's credentials and permissions. The hijacked process can expose tools to agent sessions without detection. The attack requires no user interaction and can be triggered by external content like web pages or emails. It allows unrestricted network access and remains invisible to both users and models.

Compliance Impact

This vulnerability could lead to unauthorized code execution with elevated permissions, potentially exposing sensitive data. This may violate GDPR's data protection principles (e.g., integrity, confidentiality) and HIPAA's security requirements for protected health information if exploited in healthcare environments.

Mitigation Strategies

Avoid using pnpm source-checkout deployments. Use fully qualified command paths in MCP configurations. Restrict write permissions in node_modules/.bin. Implement a unified executor-resolution layer to prevent bare-name spawns.

Chat Assistant

Ask questions about this CVE
Hi! I’m here to help you understand CVE-2026-95182. Ask me anything about the vulnerability, its impact, or mitigation strategies.
0/70

EPSS Chart