Loading vulnerability…
CVE-2026-12940 is an unauthenticated remote code execution vulnerability affecting IBM Langflow OSS versions 1.0.0 through 1.10.1. The flaw is in the MCP (Model Context Protocol) stdio launcher, specifically in src/lfx/src/lfx/base/mcp/util.py, where the DANGEROUS_ENV_VARS blocklist is incomplete and does not include the SHELLOPTS, BASHOPTS, and PS4 environment variables. This allows environment variable injection into a shell-invocation context, resulting in OS command injection and arbitrary code execution. The issue is consistent with CWE-78 because attacker-controlled input can influence command execution by the underlying shell. Successful exploitation permits remote execution of arbitrary commands without authentication.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
Repository is a compact Python proof-of-concept for CVE-2026-12940 targeting Langflow OSS <= 1.10.1. The main exploit file is exploit.py, supported by README.md, ANALYSIS.md, and EVIDENCE.txt documenting root cause and validation, plus lab/docker-compose.yml for reproducing the vulnerable environment. The exploit is real and operational: it first obtains a bearer token from GET /api/v1/auto_login when AUTO_LOGIN is enabled by default, then registers an MCP stdio server through POST /api/v2/mcp/servers/{name} using a benign allowlisted command (uvx mcp-server-fetch) but malicious environment variables SHELLOPTS=xtrace and PS4=$(command). Finally, it triggers GET /api/v2/mcp/servers?action_count=true, which causes Langflow to enumerate configured MCP servers and spawn bash -c with attacker-controlled env. Because bash expands PS4 under xtrace before executing the first traced command, arbitrary shell commands run regardless of the command/args allowlist. The exploit therefore provides remote code execution as the Langflow service user, with examples including id output redirection to /tmp/proof and outbound TCP callbacks. Repository structure: 1 Python exploit entry point, 1 YAML lab deployment file, and 3 markdown/text documentation artifacts. No external exploit framework is used.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
8 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
A critical unauthenticated remote code execution vulnerability in LangFlow caused by improper input validation in the MCP (Model Context Protocol) stdio launcher, enabling OS command injection and arbitrary code execution.
An unauthenticated remote code execution vulnerability in IBM Langflow OSS caused by environment variable injection in the MCP stdio launcher due to an incomplete dangerous environment variable blocklist.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.