CVE-2026-101861: Langflow: eval() code injection via component options

MEDIUMPoC AVAILABLE
Published September 28, 2026
CISO Take

Langflow versions before 1.12.0 pass component input options into a Literal type string that is evaluated with eval(). An authenticated user with high privileges can plant a Python object with a malicious __repr__ and get code execution on the server when a component is converted to a LangChain tool, including during custom component saves through the API. Risk is moderate: CVSS is 4.1 (high attack complexity, high privileges required), EPSS is 0.00197 (top 91% percentile), there is no public exploit, no Nuclei template, it is not in CISA KEV, and CISA SSVC says TRACK. The concern is that Langflow already has 194 other CVEs and often holds LLM API keys and connector credentials, so any code execution there gives an attacker access to those secrets. Upgrade to Langflow 1.12.0 or later, and in the meantime limit who can create or save custom components and expose the API only to trusted networks.

Sources:NVDEPSSGitHub AdvisoryATLASCISA SSVC

What is the risk?

Overall risk is MEDIUM. Exploitation needs network access, high privileges and high complexity, which lowers the likelihood. There is no public exploit, no scanner template and no KEV listing. The impact is arbitrary Python execution in the Langflow server process, though the CVSS vector rates confidentiality, integrity and availability impact as low. Real impact can be higher because Langflow servers hold LLM provider keys, vector database credentials and connector secrets. The package has a risk score of 77/100 and 194 other CVEs, which points to a recurring code-injection weakness class (CWE-94, CWE-95). Exposure is highest for multi-user or shared Langflow deployments where several people can author components.

How does the attack unfold?

Initial Access
Attacker authenticates to Langflow with a high-privilege account, such as a compromised or malicious builder.
AML.T0012
Payload Injection
Attacker places a Python object with a malicious __repr__ in a custom component's input options list and saves it through the API.
AML.T0049
Code Execution
ComponentToolkit.get_tools() interpolates the options into a Literal type string and eval() runs the attacker code.
AML.T0050
Impact
Attacker harvests LLM and database credentials and pivots to connected data stores and services.

What systems are affected?

PackageEcosystemVulnerable RangePatched
Langflowpip—No patch
156.7KPushed 3d ago18% patched~76d to patchFull package profile →

Do you use Langflow? You're affected.

How severe is it?

CVSS 3.1
4.1 / 10
EPSS
0.2%
chance of exploitation in 30 days
Higher than 9% of all CVEs
Exploitation Status
Exploit Available
Exploitation: MEDIUM
Sophistication
Moderate
Exploitation Confidence
medium
○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?

AVACPRUISCIA
AVNetwork
ACHigh
PRHigh
UINone
SUnchanged
CLow
ILow
ALow

What should I do?

1 step
  1. 1) Upgrade Langflow to 1.12.0 or later, per the vendor release notes and advisory GHSA-33p4-w7j3-33mw. 2) Until patched, restrict who can create, edit or save custom components and tools, and remove shared or low-trust accounts. 3) Do not expose the Langflow API or UI to the internet; put it behind a VPN or authenticated reverse proxy. 4) Run Langflow as an unprivileged user in a container with minimal egress and no broad secrets. 5) Rotate LLM provider keys and connector credentials if untrusted users had author access. 6) Detection: review API logs for custom component save requests with unusual options lists, and alert on unexpected child processes or outbound connections from the Langflow process.

What does CISA's SSVC say?

DecisionTrack
Exploitationnone
AutomatableNo
Technical Impactpartial

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

How is it classified?

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-Responses to identified AI risks
OWASP LLM Top 10
LLM03-Supply ChainLLM06-Excessive Agency

Frequently Asked Questions

What is CVE-2026-101861?

Langflow versions before 1.12.0 pass component input options into a Literal type string that is evaluated with eval(). An authenticated user with high privileges can plant a Python object with a malicious __repr__ and get code execution on the server when a component is converted to a LangChain tool, including during custom component saves through the API. Risk is moderate: CVSS is 4.1 (high attack complexity, high privileges required), EPSS is 0.00197 (top 91% percentile), there is no public exploit, no Nuclei template, it is not in CISA KEV, and CISA SSVC says TRACK. The concern is that Langflow already has 194 other CVEs and often holds LLM API keys and connector credentials, so any code execution there gives an attacker access to those secrets. Upgrade to Langflow 1.12.0 or later, and in the meantime limit who can create or save custom components and expose the API only to trusted networks.

Is CVE-2026-101861 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-101861, increasing the risk of exploitation.

How to fix CVE-2026-101861?

1) Upgrade Langflow to 1.12.0 or later, per the vendor release notes and advisory GHSA-33p4-w7j3-33mw. 2) Until patched, restrict who can create, edit or save custom components and tools, and remove shared or low-trust accounts. 3) Do not expose the Langflow API or UI to the internet; put it behind a VPN or authenticated reverse proxy. 4) Run Langflow as an unprivileged user in a container with minimal egress and no broad secrets. 5) Rotate LLM provider keys and connector credentials if untrusted users had author access. 6) Detection: review API logs for custom component save requests with unusual options lists, and alert on unexpected child processes or outbound connections from the Langflow process.

What systems are affected by CVE-2026-101861?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, LLM orchestration platforms, RAG pipelines, low-code AI workflow builders.

What is the CVSS score for CVE-2026-101861?

CVE-2026-101861 has a CVSS v3.1 base score of 4.1 (MEDIUM). The EPSS exploitation probability is 0.20%.

What is the AI security impact?

Affected AI Architectures

agent frameworksLLM orchestration platformsRAG pipelineslow-code AI workflow builders

MITRE ATLAS Techniques

AML.T0012Valid Accounts
AML.T0050Command and Scripting Interpreter
AML.T0053AI Agent Tool Invocation

Compliance Controls Affected

EU AI Act: Art. 15
ISO 42001: A.6.2.6
NIST AI RMF: MANAGE 2.3
OWASP LLM Top 10: LLM03, LLM06

What are the technical details?

Original Advisory

Langflow 1.0.16 before 1.12.0 and 0.0.94 before 1.12.0 contain an unsafe eval() vulnerability in schema.py that allows authenticated attackers to achieve code execution by placing a Python object with a malicious __repr__ method into component input options lists. The eval() sink is triggered when a component is converted into a LangChain tool via ComponentToolkit.get_tools(), including during custom component saves through the API, by interpolating options into a Literal type string that is passed directly to eval() without safe evaluation controls.

Exploitation Scenario

An attacker with an authenticated, high-privilege Langflow account (for example a compromised builder or a malicious insider on a shared instance) submits a custom component through the API. Its input options list contains a Python object whose __repr__ method runs attacker code. When the component is saved or converted to a tool, ComponentToolkit.get_tools() builds a Literal type string from the options and calls eval(), which invokes __repr__ and runs the code on the server. The attacker then reads environment variables with LLM and database keys, and uses them to reach the connected data stores.

Weaknesses (CWE)

CWE-94 — Improper Control of Generation of Code: The product constructs all or part of a code segment using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the syntax or behavior of the intended code segment.

  • [Architecture and Design] Refactor your program so that you do not have to dynamically generate code.
  • [Architecture and Design] Run your code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which code can be executed by your product. Examples include the Unix chroot jail and AppArmor. In general, managed code may provide some protection. This may not be a feasible solution, and it only limits the impact to the operating system; the rest of your application may still be subject to compromise. Be careful to avoid CWE-243 and other weaknesses related to jails.

Source: MITRE CWE corpus.

CVSS Vector

CVSS:3.1/AV:N/AC:H/PR:H/UI:N/S:U/C:L/I:L/A:L

Timeline

Published
September 28, 2026
Last Modified
September 30, 2026
First Seen
September 28, 2026

Related Vulnerabilities