CVE-2026-105740: Langflow: MCP Stdio command injection gives authenticated RCE

CRITICALPoC AVAILABLECISA: ATTEND

Published October 5, 2026

CISO Take

Langflow versions before 1.9.0 let any authenticated user run arbitrary OS commands on the server. They do it by registering an MCP server with the Stdio transport, because the command field is passed unvalidated to bash -c and runs as soon as the server list is fetched. The CVSS is 9.9 (network, low complexity, low privileges, scope changed), and the env field also allows LD_PRELOAD and PATH injection. EPSS is low at 0.00589 (about the 54th percentile). It is not in CISA KEV, and no public exploit or Nuclei template is known, but the flaw is trivial to abuse for anyone holding a valid account. Upgrade to Langflow 1.9.0 or later now, restrict who can create accounts and MCP servers, and rotate any API keys and secrets stored on instances that were exposed to untrusted users.

Sources:NVDEPSSGitHub AdvisoryATLAS

What is the risk?

Critical severity (CVSS 9.9). Exploitation needs only a low-privilege authenticated account, with no user interaction and low complexity. The changed scope means a compromise reaches beyond the application into the host and connected systems. Current exploitation likelihood is low to moderate: EPSS is 0.00589, there is no KEV listing, no public exploit and no scanner template. Risk is highest for multi-user or internet-facing Langflow deployments that allow open signups or share accounts. Langflow instances usually hold LLM provider keys, database credentials and vector store access, which raises the impact. Confidence in this assessment is HIGH for severity and MEDIUM for exploitation likelihood, since dependents and package telemetry are unknown.

How does the attack unfold?

Initial Access

Attacker logs in with any valid low-privilege Langflow account.

AML.T0012

Malicious Configuration

Attacker adds an MCP server using Stdio transport with a malicious command and env variables such as LD_PRELOAD.

AML.T0081

Execution

Fetching the server list triggers bash -c with the attacker's command, giving RCE on the host.

AML.T0050

Impact

Attacker steals API keys and secrets, tampers with flows and pivots into connected systems.

AML.T0049

Initial Access

Attacker logs in with any valid low-privilege Langflow account.

AML.T0012

Malicious Configuration

Attacker adds an MCP server using Stdio transport with a malicious command and env variables such as LD_PRELOAD.

AML.T0081

Execution

Fetching the server list triggers bash -c with the attacker's command, giving RCE on the host.

AML.T0050

Impact

Attacker steals API keys and secrets, tampers with flows and pivots into connected systems.

AML.T0049

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
Langflow	pip	—	No patch
156.0KPushed 6d ago18% patched~66d to patchFull package profile →

Do you use Langflow? You're affected.

How severe is it?

CVSS 3.1

9.9 / 10

EPSS

1.0%

chance of exploitation in 30 days

Higher than 64% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

✓VulnCheck KEV (exploitation reported — broader/earlier than CISA) — Oct 2026

○CISA SSVC: Public PoC

○Public PoC indexed (trickest/cve)

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AVNetwork

ACLow

PRLow

UINone

SChanged

CHigh

IHigh

AHigh

What should I do?

1 step

1) Upgrade Langflow to 1.9.0 or later (fix in PR #12290 and commit efbc4a16). 2) Until patched, restrict Langflow access to trusted users, disable open registration and put the instance behind SSO or a VPN. 3) Block or disable MCP server creation by non-admin roles where possible. 4) Run Langflow in a container with a non-root user, a read-only filesystem, minimal egress and no access to cloud metadata endpoints. 5) Rotate LLM API keys, database credentials and other secrets on any instance that was reachable by untrusted users. 6) Detect: review MCP server configurations for unexpected Stdio commands, shell metacharacters or env keys such as LD_PRELOAD and PATH. Monitor for bash -c child processes spawned by the Langflow process and for unusual outbound connections. 7) Follow the GitHub advisories GHSA-7w94-79vh-5mr2 and GHSA-w794-rj3p-xv45.

What does CISA's SSVC say?

DecisionAttend

Exploitationpoc

AutomatableNo

Technical Impacttotal

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Code ExecutionSupply ChainFrameworkAgentPluginAML.T0012- Valid Accounts AML.T0049- Exploit Public-Facing Application AML.T0050- Command and Scripting Interpreter AML.T0081- Modify AI Agent Configuration

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art. 15-Accuracy, robustness and cybersecurity

ISO 42001

A.6.2.6-AI system operation and monitoring

NIST AI RMF

MANAGE 2.3-Response to previously unknown risks and mechanisms

OWASP LLM Top 10

LLM05-Supply Chain VulnerabilitiesLLM06-Excessive Agency

Frequently Asked Questions

What is CVE-2026-105740?

Is CVE-2026-105740 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-105740, increasing the risk of exploitation.

How to fix CVE-2026-105740?

1) Upgrade Langflow to 1.9.0 or later (fix in PR #12290 and commit efbc4a16). 2) Until patched, restrict Langflow access to trusted users, disable open registration and put the instance behind SSO or a VPN. 3) Block or disable MCP server creation by non-admin roles where possible. 4) Run Langflow in a container with a non-root user, a read-only filesystem, minimal egress and no access to cloud metadata endpoints. 5) Rotate LLM API keys, database credentials and other secrets on any instance that was reachable by untrusted users. 6) Detect: review MCP server configurations for unexpected Stdio commands, shell metacharacters or env keys such as LD_PRELOAD and PATH. Monitor for bash -c child processes spawned by the Langflow process and for unusual outbound connections. 7) Follow the GitHub advisories GHSA-7w94-79vh-5mr2 and GHSA-w794-rj3p-xv45.

What systems are affected by CVE-2026-105740?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, LLM workflow builders, MCP tool integrations, RAG pipelines, multi-user AI development platforms.

What is the CVSS score for CVE-2026-105740?

CVE-2026-105740 has a CVSS v3.1 base score of 9.9 (CRITICAL). The EPSS exploitation probability is 1.05%.

What is the AI security impact?

Affected AI Architectures

agent frameworksLLM workflow buildersMCP tool integrationsRAG pipelinesmulti-user AI development platforms

MITRE ATLAS Techniques

AML.T0012Valid Accounts

AML.T0049Exploit Public-Facing Application

AML.T0050Command and Scripting Interpreter

AML.T0081Modify AI Agent Configuration

Compliance Controls Affected

EU AI Act: Art. 15

ISO 42001: A.6.2.6

NIST AI RMF: MANAGE 2.3

OWASP LLM Top 10: LLM05, LLM06

What are the technical details?

Original Advisory

Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to 1.9.0, any authenticated Langflow user can achieve Remote Code Execution (RCE) on the server by adding an MCP server with the "Stdio" transport. The user-supplied command field is passed directly to bash -c "exec {command}" with zero validation, no allowlisting, and no sandboxing. The command executes immediately when the server list is fetched. Additionally, the env field allows arbitrary environment variable injection (e.g., LD_PRELOAD, PATH override). This vulnerability is fixed in 1.9.0.

Exploitation Scenario

An attacker obtains a low-privilege Langflow account through open signup, a phished user or leaked credentials. They add a new MCP server with the Stdio transport and set the command field to a reverse shell or a credential-dumping one-liner. They may also set env variables such as LD_PRELOAD. When the server list is fetched, Langflow runs the command through bash -c on the host. The attacker then reads LLM provider keys and database secrets from the environment, modifies agent flows to persist, and moves laterally into data stores and cloud resources connected to the AI platform.

Weaknesses (CWE)

CWE-78Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')Primary CWE-78Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection')

CWE-78 — Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection'): The product constructs all or part of an OS command using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the intended OS command when it is sent to a downstream component.

[Architecture and Design] If at all possible, use library calls rather than external processes to recreate the desired functionality.
[Architecture and Design, Operation] Run the code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which files can be accessed in a particular directory or which commands can be executed by the software. OS-level examples include the Unix chroot jail, AppArmor, and SELinux. In general, managed code may provide some protection. For example, java.io.FilePermission in the Java SecurityManager allows the software to specify restrictions on file operations. This may not be a feasible solution, and it only limits the impact to the operating system; the rest of the application may still be subject to compromise. Be careful to avoid CWE-243 and other weaknesses related to jails.

Source: MITRE CWE corpus.