CVE-2026-27966
📋 TL;DR
This vulnerability in Langflow's CSV Agent node allows attackers to execute arbitrary Python and OS commands on the server via prompt injection, leading to full Remote Code Execution (RCE). It affects all Langflow deployments using versions before 1.8.0 where the CSV Agent node is enabled. The issue stems from hardcoded dangerous code execution settings that expose LangChain's Python REPL tool.
💻 Affected Systems
- Langflow
📦 What is this software?
Langflow by Langflow
⚠️ Risk & Real-World Impact
Worst Case
Complete server compromise allowing data theft, lateral movement, persistence establishment, and full system control.
Likely Case
Attacker gains shell access to execute commands, potentially stealing sensitive data, installing malware, or disrupting services.
If Mitigated
Limited impact if proper network segmentation, least privilege, and monitoring are in place, though RCE still poses significant risk.
🎯 Exploit Status
Exploitation requires prompt injection through the CSV Agent node interface, which typically requires some level of access to the Langflow application.
🛠️ Fix & Mitigation
✅ Official Fix
Patch Version: 1.8.0
Vendor Advisory: https://github.com/langflow-ai/langflow/security/advisories/GHSA-3645-fxcv-hqr4
Restart Required: Yes
Instructions:
1. Update Langflow to version 1.8.0 or later using pip: 'pip install --upgrade langflow==1.8.0' 2. Restart the Langflow service or application 3. Verify the update with 'langflow --version'
🔧 Temporary Workarounds
Disable CSV Agent Node
allTemporarily disable or remove the CSV Agent node from all workflows to prevent exploitation.
Network Isolation
allRestrict network access to Langflow instances to trusted users only using firewall rules.
🧯 If You Can't Patch
- Implement strict input validation and sanitization for all user inputs to the CSV Agent node
- Deploy Langflow in a containerized environment with minimal privileges and network restrictions
🔍 How to Verify
Check if Vulnerable:
Check if Langflow version is below 1.8.0 and if CSV Agent node is present in any workflows.
Check Version:
langflow --version
Verify Fix Applied:
Confirm Langflow version is 1.8.0 or higher and review the commit d8c6480daa17b2f2af0b5470cdf5c3d28dc9e508 has been applied.
📡 Detection & Monitoring
Log Indicators:
- Unusual Python execution patterns
- Suspicious OS commands in Langflow logs
- Multiple failed prompt injection attempts
Network Indicators:
- Unexpected outbound connections from Langflow server
- Traffic to known malicious IPs
SIEM Query:
source="langflow.logs" AND ("python_repl_ast" OR "allow_dangerous_code" OR "os.system" OR "subprocess")