SSRF + LLM API key exfiltration via user-controlled basepath / baseUrl in 8 chat-model nodes
CVE-2026-91935 · GHSA-hx55-h48h-7rw9 · published 2026-08-31 · updated 2026-09-20
- Severity
- Critical
- CVSS score
- 9.4
- Chance of exploitation (EPSS)
- Not scored
- Fixed in
- 3.1.4
- Exploitation Public proof of concept
- Automatable No
- Technical impact Total
- CISA SSVC
1. SSRF with full response read. Any workspace member with chatflow-edit permission can cause the Flowise server to issue HTTP POST requests to any reachable host — cloud metadata (169.254.169.254, metadata.google.internal), internal services (postgres:5432, redis:6379, internal admin dashboards), link-local, IPv6 loopback. Response is surfaced as the LLM chat output, so this is a read-anywhere primitive, not blind. 2. Cloud IAM credential theft on EC2/ECS/EKS/GKE. returns temporary AWS credentials. Flowise has no IMDS protection for these chat-model nodes. 3. LLM provider API key exfiltration. …
Upgrade flowise to 3.1.4 or later, and flowise-components to 3.1.4 or later. If you start it from an MCP config with npx or uvx, pin that version so an old cached copy is not used.
1. Reject private/loopback/link-local URLs in every chat-model node init() before constructing the LangChain client. The simplest fix is a shared helper that calls checkDenyList(basePath) from httpSecurity.ts at node-init time: 2. Configure LangChain's httpAgent to go through the Flowise SSRF proxy. The ChatOpenAI constructor accepts configuration.fetch — route it through a secureFetch-wrapped function that re-checks every URL, including redirects: 3. Apply the same fix to ChatOllama, ChatLocalAI, ChatLitellm, ChatNvdiaNIM, ChatNemoGuardrails, ChatCloudflareWorkersAI, ChatGoogleGenerativeAI. Each has its own init() that constructs a LangChain client; each needs the same guard. …
| Package | Registry | Affected | Fixed in |
|---|---|---|---|
| flowise | npm | <= 3.1.3 | 3.1.4 |
| flowise-components | npm | <= 3.1.3 | 3.1.4 |
- CriticalMLflow: Unauthenticated full-read SSRF in webhook delivery: _validate_webhook_url bypassed via unvalidated HTTP redirects (and DNS rebindingmlflowExploited
- Critical@zereight/mcp-gitlab Vulnerable to Server-Side Request Forgery@zereight/mcp-gitlab
- CriticalLocalAI POST /models/apply permits unauthenticated server-side request forgery through gallery URLsgithub.com/mudler/LocalAI
- CriticalFastMCP OpenAPI Provider has an SSRF & Path Traversal Vulnerabilityfastmcp
- CriticalBentoML SSRF Vulnerability in File Upload Processingbentoml
- Criticallobe-chat `/api/proxy` endpoint Server-Side Request Forgery vulnerability@lobehub/chat
- CriticalFlowise NodeVM sandbox escape via puppeteer allowlist - authenticated RCE and arbitrary file read via Chromiumflowise, flowise-components
- CriticalFlowise Prompt Injection to RCE and SSRF via CSV/Airtable Agent Python Validator Bypassflowise, flowise-components
- CriticalCross-IdP account takeover via email-only SSO matchingflowise
- CriticalINVITED users auto-promoted via SSO email match (invite-token bypass)flowise
- CriticalFlowise RCE via SQL Database Chain Nodeflowise, flowise-components
- CriticalFlowise Custom MCP npx package execution leads to authenticated server-side command executionflowise, flowise-components
The impact, patch and workaround text comes from the advisory, shortened (published by Flowise). Proof-of-concept and reproduction sections and code are left out; the original advisory has them.