CVE-2025-1796: Langgenius Dify Auth Bypass Vulnerability

CVE-2025-1796 Overview

CVE-2025-1796 is an account takeover vulnerability in langgenius/dify version 0.10.1, an open-source LLM application development platform. The flaw stems from the use of random.randint, a non-cryptographic pseudo-random number generator (PRNG), to generate password reset codes. An authenticated attacker with access to workflow tools can extract PRNG outputs, reconstruct the internal state, and predict future reset codes. This enables takeover of any account, including administrator accounts, leading to full application compromise. The weakness is classified under CWE-338: Use of Cryptographically Weak Pseudo-Random Number Generator.

Critical Impact
An authenticated attacker can predict password reset codes and seize control of administrator accounts, resulting in full compromise of the dify instance.

Affected Products

langgenius/dify version 0.10.1
Node.js-based dify deployments matching the affected CPE
Self-hosted dify instances exposing workflow tools to low-privilege users

Discovery Timeline

2025-03-20 - CVE-2025-1796 published to NVD
2026-06-17 - Last updated in NVD database

Technical Details for CVE-2025-1796

Vulnerability Analysis

The vulnerability resides in dify's password reset workflow. The application calls Python's random.randint to generate the numeric reset code sent to a user's email. Python's random module implements a Mersenne Twister algorithm, which is deterministic and reversible once enough sequential outputs are observed. The module is documented as unsuitable for security or cryptographic purposes.

Because dify exposes workflow tools that surface PRNG-derived values to authenticated users, an attacker can collect successive outputs from the same PRNG instance. With 624 consecutive 32-bit outputs, the full Mersenne Twister internal state can be reconstructed, allowing precise prediction of all subsequent values, including reset codes issued to other users.

The attacker then triggers a password reset for a target administrator account and supplies the predicted code to reset the password and seize the account.

Root Cause

The root cause is the selection of a non-cryptographic PRNG for security-sensitive token generation. Secure reset codes require a cryptographically secure source such as Python's secrets module or os.urandom. Using random.randint violates secure design principles documented under CWE-338.

Attack Vector

The attack requires network access and low-privilege authentication to a dify instance. The attacker performs the following steps in prose:

Authenticate to dify and access workflow tools that expose values derived from random.
Repeatedly invoke the workflow to harvest 624 sequential 32-bit PRNG outputs.
Reconstruct the Mersenne Twister state using publicly available untwister tooling.
Predict the next reset code value the application will produce.
Trigger a password reset for an administrator email address.
Submit the predicted code to set a new password and log in as the administrator.

Further technical details are available in the Huntr Bounty Listing.

Detection Methods for CVE-2025-1796

Indicators of Compromise

Repeated invocations of workflow tools by a single low-privilege account in a short time window, consistent with PRNG output harvesting.
Password reset requests for administrator or privileged accounts followed by successful authentication from an unfamiliar source IP.
Unexpected password changes on administrator accounts without a corresponding helpdesk ticket or user request.

Detection Strategies

Audit dify application logs for high-frequency workflow execution by individual users, especially workflows that return random or numeric values.
Correlate password reset events with account login events from new IP addresses or user agents within minutes of the reset.
Alert on administrator account password changes that lack a corresponding authenticated session from the legitimate user.

Monitoring Recommendations

Forward dify authentication, password reset, and workflow execution logs to a centralized logging platform for retention and correlation.
Establish a baseline for normal workflow execution rates per user and alert on deviations above that baseline.
Monitor for new administrator session creation immediately after a password reset and require manual verification.

How to Mitigate CVE-2025-1796

Immediate Actions Required

Upgrade dify to a release later than 0.10.1 that replaces random.randint with a cryptographically secure generator such as secrets.randbelow.
Force a password reset for all users, prioritizing administrator and privileged accounts, after upgrading.
Restrict access to workflow tools to trusted users until the patched version is deployed.
Review recent password reset events and administrator logins for evidence of exploitation.

Patch Information

No vendor advisory URL is listed in the NVD record at the time of writing. Refer to the Huntr Bounty Listing and the langgenius/dify project repository for the fixed release and commit details. Apply the upgrade in a staged environment before production rollout.

Workarounds

Disable self-service password reset and require administrator-driven password resets until the patch is applied.
Remove or restrict workflow tools that expose PRNG-derived values to low-privilege users.
Place the dify management interface behind a VPN or zero-trust proxy to limit attacker reach to the password reset endpoint.

bash

# Configuration example: restrict dify access at the reverse proxy layer
# Example nginx snippet limiting password reset and workflow endpoints to trusted CIDRs
location ~ ^/(console/api/forgot-password|console/api/workflows) {
    allow 10.0.0.0/8;
    deny all;
    proxy_pass http://dify_backend;
}