CVE-2023-34541 — CRITICAL (CVSS 9.8) AI Security Vulnerability

CISO Take

Any LangChain deployment at version 0.0.171 or earlier using load_prompt is critically exposed to unauthenticated remote code execution — no user interaction required. Upgrade to a patched version immediately; if patching is blocked, disable all load_prompt calls or restrict them to fully trusted, internal sources. Given LangChain's prevalence in AI agent and RAG architectures, assume blast radius extends to the host system and downstream model serving infrastructure.

Risk Assessment

Severity is maximum practical risk: CVSS 9.8 with network-accessible attack vector, low complexity, no privileges, no user interaction, and full C/I/A impact. LangChain 0.0.171 was widely deployed at publication time across production AI pipelines, making the exposed surface area large. The vulnerability requires only the ability to supply an attacker-controlled URL or file path to load_prompt, which is a common pattern in agentic and configuration-driven LangChain applications. No weaponization barrier exists — proof-of-concept exploits were publicly linked in the original GitHub issue.

Affected Systems

Package	Ecosystem	Vulnerable Range	Patched
langchain	pip	—	No patch
136.3K OpenSSF 6.4 2.6K dependents Pushed today 17% patched ~256d to patch Full package profile →

Do you use langchain? You're affected.

Severity & Risk

CVSS 3.1

9.8 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 37% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

○ Public PoC indexed (trickest/cve)

Composite signal derived from CISA KEV, CISA SSVC, EPSS, trickest/cve, and Nuclei templates.

Attack Surface

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I High

A High

Recommended Action

6 steps

Patch: Upgrade LangChain to any version above 0.0.171 immediately.
Inventory: Search codebase for all calls to load_prompt and document their input sources.
Workaround (if patching delayed): Remove or gate load_prompt calls behind strict allowlists of trusted, internal URIs only — never accept user-supplied or externally-sourced paths.
Containment: Run LangChain services with least-privilege OS accounts, restrict outbound network from AI servers, and isolate with network segmentation.
Detection: Monitor for anomalous subprocess spawning, unexpected outbound connections, or file reads from unusual paths in processes hosting LangChain.
Secrets audit: Rotate any API keys, model service credentials, or database passwords accessible from the compromised process scope.

CISA SSVC Assessment

Decision Attend

Exploitation poc

Automatable Yes

Technical Impact total

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

Classification

Code Execution Supply Chain Data Extraction Framework Agent AML.T0010.001 - AI Software AML.T0025 - Exfiltration via Cyber Means AML.T0049 - Exploit Public-Facing Application AML.T0050 - Command and Scripting Interpreter AML.T0072 - Reverse Shell

Compliance Impact

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.8.3 - AI system security

NIST AI RMF

GV-6.1 - Policies and procedures for AI risk and impact MS-2.5 - Practices and personnel for AI risk management

OWASP LLM Top 10

LLM05 - Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2023-34541?

Any LangChain deployment at version 0.0.171 or earlier using load_prompt is critically exposed to unauthenticated remote code execution — no user interaction required. Upgrade to a patched version immediately; if patching is blocked, disable all load_prompt calls or restrict them to fully trusted, internal sources. Given LangChain's prevalence in AI agent and RAG architectures, assume blast radius extends to the host system and downstream model serving infrastructure.

Is CVE-2023-34541 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2023-34541, increasing the risk of exploitation.

How to fix CVE-2023-34541?

1. Patch: Upgrade LangChain to any version above 0.0.171 immediately. 2. Inventory: Search codebase for all calls to load_prompt and document their input sources. 3. Workaround (if patching delayed): Remove or gate load_prompt calls behind strict allowlists of trusted, internal URIs only — never accept user-supplied or externally-sourced paths. 4. Containment: Run LangChain services with least-privilege OS accounts, restrict outbound network from AI servers, and isolate with network segmentation. 5. Detection: Monitor for anomalous subprocess spawning, unexpected outbound connections, or file reads from unusual paths in processes hosting LangChain. 6. Secrets audit: Rotate any API keys, model service credentials, or database passwords accessible from the compromised process scope.

What systems are affected by CVE-2023-34541?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, RAG pipelines, LLM application backends, model serving, AI orchestration platforms.

What is the CVSS score for CVE-2023-34541?

CVE-2023-34541 has a CVSS v3.1 base score of 9.8 (CRITICAL). The EPSS exploitation probability is 0.17%.

Technical Details

NVD Description

Langchain 0.0.171 is vulnerable to Arbitrary code execution in load_prompt.

Exploitation Scenario

An adversary hosts a malicious YAML prompt template at an attacker-controlled URL. The file embeds a Python deserialization payload (e.g., via PyYAML's !!python/object constructor or equivalent). The attacker sends a request to a LangChain-backed application endpoint that accepts a prompt configuration path — common in agent orchestration UIs, no-code AI builders, or misconfigured API endpoints. When the application calls load_prompt with the attacker URL, LangChain fetches and deserializes the file, executing the embedded payload. The attacker receives a reverse shell running as the LangChain service account, gains access to environment variables containing LLM API keys and database credentials, and can pivot to the vector store, model registry, or cloud provider APIs connected to the AI pipeline.