$RodHat_/search
MOTD

Flowise forgot-password hands you the reset token, and Metasploit now has a module

Published by

This article may contain Amazon affiliate links. I may earn a commission if you buy through them. Disclosure.

Editorial card for a Flowise password reset vulnerability
Editorial card: RodHat

The Rapid7 Metasploit wrapup dated November 6 lists a new auxiliary module, admin/http/flowise_auth_bypass_cve_2025_58434, for an account takeover in Flowise 3.0.5 and earlier. The CVE is from September 2025. NVD published it September 12 at CVSS 9.8 (AV:N/AC:L/PR:N/UI:N). It took a year and change to get a point-and-click module, which is about the usual interval between “patched” and “somebody’s intern runs it against the whole internet.”

The bug is almost insulting. Per the GitHub advisory, POST /api/v1/account/forgot-password takes an email address and, instead of just sending mail, answers with the user record: ID, name, email, hashed credential, status, timestamps, and a live tempToken with its expiry. You then hand that token straight to /api/v1/account/reset-password with a new password. No email round trip, no click, no victim. All you need is an address, and the advisory notes that’s often guessable. Admin accounts included.

I have seen this exact bug in homegrown PHP in 2003. The whole point of a reset flow is that the secret travels over a channel the attacker doesn’t control. Return it in the HTTP response and you’ve built a “log in as anyone” endpoint with extra steps. It also leaked the hashed credential, which is a bonus nobody asked for.

The fix, commit 9e178d6 in 3.0.6, is small. It adds a sanitizeUser() helper that deletes credential, tempToken and tokenExpiry from the user object, and wraps the return values in the account and user services with it. That’s what the diff shows; the advisory also recommends a generic success message regardless of input, so the endpoint stops confirming which emails exist. I’d want to see that last part confirmed in the release before I called enumeration closed. Deleting fields from an object right before returning it is a blocklist, and blocklists rot the day somebody adds a column.

What to do:

  1. Find every Flowise instance you run, including the one a colleague spun up “just to try it.” The advisory says self-hosted and the cloud service were both affected.
  2. Anything at or below 3.0.5 gets upgraded to 3.0.6 or later. Check the version, don’t trust the container tag you remember.
  3. If it was ever reachable from outside, assume reset requests were abused. Review auth logs for password resets nobody asked for, and rotate credentials and any API keys or model-provider keys stored in those flows.
  4. Put it behind a VPN or an authenticating reverse proxy. A tool that holds your LLM provider keys does not belong on a public IP.

One more thing I noticed: the advisory and commit pages now say the repository was archived by its owner on August 13, 2026 and is read-only. If you are running Flowise and relying on upstream for the next one of these, go and work out what that means for you before the next module lands. I’m not going to guess at why.

Same pattern as the Kafka JWT alg: none bypass: the auth path is the product, and it got written last. More of this sort of thing lives in the kernel and security hub. Go check your versions.

Sources

  1. CVE-2025-58434 (NIST National Vulnerability Database)
  2. Critical: Unauthenticated Password Reset Token Disclosure Leading to Account Takeover in Flowise Cloud and Local Deployments (GHSA-wgpv-6j63-x5ph) (FlowiseAI)
  3. Secure password reset endpoints (#5167) (FlowiseAI)
  4. Metasploit Wrapup: Unauth RCE modules in Network Perimeter Devices and much more (Rapid7)