CVE-2025-34291: Langflow Permissive CORS & Account Takeover to RCE Chain
HERMES THREAT SCORE & CISA KEV ACTIVELY WEAPONIZED RISK
Target:Langflow Visual AI Orchestration Framework (versions jusqu'à 1.6.9) Cataloged in the CISA Known Exploited Vulnerabilities (KEV) catalog on May 21, 2026. While rated CVSS 9.4 due to required user interaction (UI:R), Hermes assigns an EXTREME HTS score of 98 and an EXTREME Agentic Security Score of 95. In production enterprise environments, Langflow orchestrates production AI agents and connects directly to databases, vector indexes, and cloud APIs. Chaining permissive CORS with SameSite=None refresh cookies enables drive-by session hijacking, leading directly to arbitrary remote code execution on the backend server.
CVE-2025-34291: Langflow Permissive CORS & Account Takeover to RCE ChainVULNERABILITY
Visual framework and multi-agent development environment for building, evaluating, and deploying conversational AI pipelines.
🔍 Why is this related? (Evidence & Provenance)
“Confirmed security vulnerability in Langflow Visual AI Builder documented in Hermes dossier.”
- [vulnerability_report]
- [government_confirmation]CISA verified active exploitation in the wild and mandated federal remediation deadline in KEV entry. — Source: Cybersecurity & Infrastructure Security Agency (CISA): CISA Adds CVE-2026-59822 to Known Exploited Vulnerabilities Catalog (Reliability: VERY_HIGH)
Adversarial subversion of structured tool execution arguments (SQL, Shell, Filepath) passed from an LLM agent to host OS tools or MCP endpoints.
🔍 Why is this related? (Evidence & Provenance)
“CVE-2025-34291 weaponizes the agentic attack pattern formalized under AAP-003.”
- [technical_analysis]Pillar Security demonstrated that executing export BASH_ENV in Auto-Run causes bash to source hostile payloads upon subsequent commands. — Source: Pillar Security Research: Bypassing Cursor Auto-Run: When Shell Built-ins Lead to Host RCE (Reliability: HIGH)
1. Identification & Threat Intel Tracking
Section titled “1. Identification & Threat Intel Tracking”| Parameter | Technical Specification | Operational Impact |
|---|---|---|
| CVE Identifier | CVE-2025-34291 | Official MITRE / NVD entry |
| CISA KEV Status | ACTIVE EXPLOITATION CONFIRMED | Cataloged on May 21, 2026 |
| Associated CWEs | CWE-346: Origin Validation ErrorCWE-94: Code Injection | Drive-by admin account takeover & RCE |
| CVSS v3.1 Score | 9.4 CRITICAL | CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H |
| EPSS Probability | 0.978 (99.9th percentile) | Widespread weaponized threat actor utilization |
| Hermes Threat Score (HTS) | 98 / 100 (EXTREME) | Confirmed active exploitation in production |
| Hermes Agentic Score (HASS) | 95 / 100 (EXTREME) | Total compromise of enterprise AI agent fleet |
| Affected Software | langflow:langflow | Versions jusqu'à 1.6.9 |
| Fixed Version | 1.7.0 | Strict CORS whitelisting & SameSite=Lax cookies |
2. Technical Attack Chain: From Permissive CORS to Server Takeover
Section titled “2. Technical Attack Chain: From Permissive CORS to Server Takeover”The vulnerability represents a textbook example of multi-stage web boundary collapse:
graph TD Admin["Authenticated Langflow Administrator"] -->|1. Browses untrusted page| AttackerWeb["Attacker Malicious Website"] AttackerWeb -->|2. Issues Cross-Origin Fetch with credentials| LangflowAPI["Langflow Backend (:7860)"] LangflowAPI -->|3. Reflects CORS + sends SameSite=None Cookie| AttackerWeb AttackerWeb -->|4. Steals Session Token & Submits Python Component| RCE["Arbitrary Code Execution on Server"] RCE -->|5. Exfiltrates Vector DB Secrets & Model Keys| TotalLoss["Full Cloud & AI Asset Compromise"]Technical Root Cause
Section titled “Technical Root Cause”- The CORS Flaw: In Langflow’s FastAPI middleware configuration, the CORS handler permitted wildcard origins while concurrently accepting credentials:
app.add_middleware(CORSMiddleware,allow_origins=["*"],allow_credentials=True,allow_methods=["*"],allow_headers=["*"],)
- SameSite=None Cookie Setting: Langflow stored the user’s refresh token inside an HTTP cookie flagged with
SameSite=Nonewithout appropriate domain restrictions. - Session Theft: When an administrator visited an attacker-controlled web page in another tab, JavaScript on the page issued credentialed cross-origin requests to the victim’s Langflow server. The browser automatically appended the session cookies.
- Elevation to RCE: With administrative session tokens in hand, the attacker invoked Langflow’s custom component APIs to deploy and execute a weaponized Python payload, securing root access to the hosting server.
3. Detection & Forensic Indicators
Section titled “3. Detection & Forensic Indicators”Network & Access Log Patterns
Section titled “Network & Access Log Patterns”- HTTP Access Logs: Look for
POST /api/v1/login/refreshandPOST /api/v1/custom_componentrequests carrying unexpected or untrusted HTTPOriginheaders. - Process Creation: Spawning of interactive shells (
/bin/sh,/bin/bash,powershell) from thelangfloworpythonparent process.
4. Remediation & Hardening Roadmap
Section titled “4. Remediation & Hardening Roadmap”Upgrade Langflow to version 1.7.0 or later immediately:
pip install langflow>=1.7.0 --upgradeExplicitly configure the allowed CORS origins in production deployment settings:
export LANGFLOW_CORS_ALLOW_ORIGINS="https://langflow.corp.internal"export LANGFLOW_AUTO_LOGIN=falseDeploy the automated SOAR containment playbook to immediately isolate compromised Langflow pods and revoke connected OpenAI/Anthropic API keys.