CVE-2026-8183: Langflow path traversal exposes

CISO Take

A directory traversal flaw in IBM Langflow OSS (versions 1.0.0 through 1.10.3) lets a low-privileged, authenticated attacker send a URL containing "../" sequences to read arbitrary files on the host running the Langflow server. Langflow is a widely deployed low-code framework for building LLM agent workflows, and its flow definitions and environment configuration commonly hold API keys for OpenAI, Anthropic, and other model providers alongside database and vector-store credentials — exactly the kind of secrets this bug can leak. There is no evidence of active exploitation (absent from CISA KEV, CISA SSVC rates it TRACK) and EPSS sits at 0.37% (69th percentile), so this is not a fire-drill, but the network attack vector, low attack complexity, and no user-interaction requirement (CVSS 7.7, scope changed) make it trivial to weaponize once an attacker holds any low-privileged foothold in the instance. Patch to the fixed Langflow release referenced in IBM's advisory immediately, and rotate any LLM/API credentials or database secrets that lived in the environment or flow files of a vulnerable, internet-reachable instance as a precaution.

Sources: NVD EPSS ATLAS ibm.com

What is the risk?

Rated HIGH (CVSS 7.7) with a network attack vector, low complexity, and no user interaction — but confidentiality-only impact (C:H/I:N/A:N) and a requirement for low privileges (PR:L), meaning the attacker needs some baseline authenticated access before traversal can be abused. No public exploit code, no Nuclei template, and no CISA KEV listing exist yet, and EPSS (0.37%, 69th percentile) suggests exploitation isn't currently widespread. The real risk driver isn't the technique itself (directory traversal is well-understood and trivial to automate) but the blast radius: Langflow instances routinely store LLM provider keys and downstream integration credentials, so a successful read can cascade into credential theft well beyond the vulnerable host.

How does the attack unfold?

Recon

Attacker identifies a reachable Langflow instance and obtains or already holds a low-privileged authenticated foothold.

AML.T0006

Exploitation

Attacker sends an HTTP request with "../" traversal sequences to a Langflow endpoint to escape the intended file directory.

AML.T0049

Credential harvest

Attacker reads environment files or flow configs exposing LLM API keys and database credentials stored on the host.

AML.T0055

Impact

Stolen credentials are reused to run inference on the victim's billing account, exfiltrate RAG-connected data, or pivot into integrated systems.

AML.T0024

Recon

Attacker identifies a reachable Langflow instance and obtains or already holds a low-privileged authenticated foothold.

AML.T0006

Exploitation

Attacker sends an HTTP request with "../" traversal sequences to a Langflow endpoint to escape the intended file directory.

AML.T0049

Credential harvest

Attacker reads environment files or flow configs exposing LLM API keys and database credentials stored on the host.

AML.T0055

Impact

Stolen credentials are reused to run inference on the victim's billing account, exfiltrate RAG-connected data, or pivot into integrated systems.

AML.T0024

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
Langflow	pip	—	No patch
155.0K Pushed 2d ago 17% patched ~70d to patch Full package profile →

Do you use Langflow? You're affected.

How severe is it?

CVSS 3.1

7.7 / 10

EPSS

0.4%

chance of exploitation in 30 days

Higher than 31% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AV Network

AC Low

PR Low

UI None

S Changed

C High

I None

A None

What should I do?

1 step

1) Upgrade Langflow past 1.10.3 to the vendor-fixed release referenced in IBM's advisory (support.ibm.com/pages/node/7282647) as soon as possible. 2) If immediate patching isn't feasible, remove direct internet exposure of the Langflow UI/API — place it behind a VPN or authenticating reverse proxy. 3) Rotate LLM provider API keys, database credentials, and any other secrets stored in environment files or flow configs on internet-facing Langflow instances, treating them as potentially compromised. 4) Add a compensating WAF/reverse-proxy rule blocking encoded and raw "../" traversal sequences in request paths. 5) Audit access logs for traversal patterns (../, %2e%2e%2f, double-encoded variants) against Langflow endpoints to detect prior exploitation attempts. 6) Since exploitation requires low privileges, tighten least-privilege on Langflow user/API-key accounts to shrink the pool of accounts that could be used as a launchpad.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact partial

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

How is it classified?

Data Extraction Privacy Violation Framework AML.T0037 - Data from Local System AML.T0049 - Exploit Public-Facing Application AML.T0055 - Unsecured Credentials

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, Robustness and Cybersecurity

ISO 42001

A.6.2.2 - AI system security

NIST AI RMF

MEASURE 2.7 - AI system security and resilience are evaluated and documented

Frequently Asked Questions

What is CVE-2026-8183?

A directory traversal flaw in IBM Langflow OSS (versions 1.0.0 through 1.10.3) lets a low-privileged, authenticated attacker send a URL containing "../" sequences to read arbitrary files on the host running the Langflow server. Langflow is a widely deployed low-code framework for building LLM agent workflows, and its flow definitions and environment configuration commonly hold API keys for OpenAI, Anthropic, and other model providers alongside database and vector-store credentials — exactly the kind of secrets this bug can leak. There is no evidence of active exploitation (absent from CISA KEV, CISA SSVC rates it TRACK) and EPSS sits at 0.37% (69th percentile), so this is not a fire-drill, but the network attack vector, low attack complexity, and no user-interaction requirement (CVSS 7.7, scope changed) make it trivial to weaponize once an attacker holds any low-privileged foothold in the instance. Patch to the fixed Langflow release referenced in IBM's advisory immediately, and rotate any LLM/API credentials or database secrets that lived in the environment or flow files of a vulnerable, internet-reachable instance as a precaution.

Is CVE-2026-8183 actively exploited?

No confirmed active exploitation of CVE-2026-8183 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-8183?

1) Upgrade Langflow past 1.10.3 to the vendor-fixed release referenced in IBM's advisory (support.ibm.com/pages/node/7282647) as soon as possible. 2) If immediate patching isn't feasible, remove direct internet exposure of the Langflow UI/API — place it behind a VPN or authenticating reverse proxy. 3) Rotate LLM provider API keys, database credentials, and any other secrets stored in environment files or flow configs on internet-facing Langflow instances, treating them as potentially compromised. 4) Add a compensating WAF/reverse-proxy rule blocking encoded and raw "../" traversal sequences in request paths. 5) Audit access logs for traversal patterns (`../`, `%2e%2e%2f`, double-encoded variants) against Langflow endpoints to detect prior exploitation attempts. 6) Since exploitation requires low privileges, tighten least-privilege on Langflow user/API-key accounts to shrink the pool of accounts that could be used as a launchpad.

What systems are affected by CVE-2026-8183?

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

What is the CVSS score for CVE-2026-8183?

CVE-2026-8183 has a CVSS v3.1 base score of 7.7 (HIGH). The EPSS exploitation probability is 0.37%.

What is the AI security impact?

Affected AI Architectures

agent frameworksRAG pipelinesLLM orchestration/workflow builders

MITRE ATLAS Techniques

AML.T0037 Data from Local System

AML.T0049 Exploit Public-Facing Application

AML.T0055 Unsecured Credentials

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.6.2.2

NIST AI RMF: MEASURE 2.7

What are the technical details?

Original Advisory

IBM Langflow OSS 1.0.0 through 1.10.3, 1.0.0 through 1.10.3, 1.0.0 through 1.10.3, 1.0.0 through 1.10.3, 1.0.0 through 1.10.3, and 1.0.0 through 1.10.3 could allow a remote attacker to traverse directories on the system. An attacker could send a specially crafted URL request containing "dot dot " sequences ( /.. /) to v i ew arbitrary files on the system.

Exploitation Scenario

An attacker who has obtained low-privileged access to a Langflow deployment — for example, a limited API key on a shared or multi-tenant instance — crafts an HTTP request to a file-serving endpoint using dot-dot-slash sequences (e.g. `/../../.env` or `/../../app/secrets.json`) to escape the intended directory. The traversal returns the environment file wiring the deployed flows to OpenAI/Anthropic API keys and a vector database connection string. The attacker exfiltrates these credentials and reuses them to run inference under the victim's billing account (cost harvesting), pull proprietary data out of the connected RAG knowledge base, or pivot laterally using the leaked database/service credentials.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal') Primary CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.