CVE-2026-79483: FastGPT NoSQL Injection Vulnerability

CVE-2026-79483 Overview

CVE-2026-79483 is a NoSQL injection vulnerability in FastGPT Community Edition versions 4.10.0 through 4.14.0. The flaw resides in the POST /api/core/chat/getHistories endpoint, which fails to sanitize JSON input before passing it to the underlying database query. An unauthenticated attacker can inject malicious NoSQL operators through crafted JSON payloads to bypass authorization checks. Successful exploitation exposes chat history titles belonging to every user on the platform. The weakness is classified under CWE-943: Improper Neutralization of Special Elements in Data Query Logic.

Critical Impact
Unauthenticated attackers can read chat history titles of all FastGPT users across the platform by bypassing authorization with crafted NoSQL operators.

Affected Products

FastGPT Community Edition 4.10.0
FastGPT Community Edition versions 4.11.x through 4.13.x
FastGPT Community Edition 4.14.0

Discovery Timeline

2026-08-31 - CVE-2026-79483 published to NVD
2026-09-01 - Last updated in NVD database

Technical Details for CVE-2026-79483

Vulnerability Analysis

FastGPT is an open-source large language model (LLM) platform maintained by labring. The Community Edition exposes a REST endpoint at /api/core/chat/getHistories that returns chat history metadata. In affected releases, the endpoint accepts user-supplied JSON fields and forwards them directly into the MongoDB query used to authorize and retrieve records. Because the query builder does not validate object types or reject query operators, attackers can substitute expected scalar values with NoSQL operator objects such as $ne, $gt, or $regex. This transforms an equality check on ownership into a broad match that returns records for arbitrary users. The vulnerability is network-reachable and requires no authentication, making it trivial to weaponize against exposed FastGPT deployments.

Root Cause

The root cause is missing type enforcement and operator neutralization on JSON input before it reaches the MongoDB query layer. The endpoint treats untrusted object structures as trusted query fragments, violating the separation between data and query logic that CWE-943 describes.

Attack Vector

An attacker sends a crafted HTTP POST request to /api/core/chat/getHistories with a JSON body that replaces one or more expected string fields with a NoSQL operator object. The manipulated operator forces the ownership predicate to evaluate as true across all records, and the server responds with chat history titles belonging to every user. No credentials, tokens, or user interaction are required. Proof-of-concept details are available in the public GitHub PoC repository.

No verified exploit code is reproduced here. Refer to the linked repository and the FastGPT project source for technical specifics.

Detection Methods for CVE-2026-79483

Indicators of Compromise

Requests to /api/core/chat/getHistories containing JSON operators such as $ne, $gt, $regex, or $where in the request body.
Anomalous response sizes from the getHistories endpoint returning records across multiple teamId or tmbId values in a single request.
Unauthenticated HTTP POST traffic to FastGPT chat APIs from external or unexpected source addresses.

Detection Strategies

Inspect application and reverse-proxy logs for POST bodies to FastGPT chat endpoints containing {"$ operator patterns.
Deploy Web Application Firewall (WAF) rules that reject JSON request bodies where expected string fields resolve to object types.
Correlate access logs against MongoDB slow-query or profiler output to spot queries returning oversized result sets from the chatHistories collection.

Monitoring Recommendations

Alert on repeated 200-response POSTs to /api/core/chat/getHistories from unauthenticated sessions or missing bearer tokens.
Baseline normal getHistories response sizes per tenant and flag statistical outliers.
Ingest FastGPT reverse-proxy and MongoDB audit logs into a centralized analytics platform for retroactive hunting.

How to Mitigate CVE-2026-79483

Immediate Actions Required

Restrict network access to FastGPT Community Edition instances so the API is not reachable from the public internet until patched.
Place FastGPT behind an authenticating reverse proxy that rejects requests missing valid session credentials.
Deploy a WAF rule that blocks JSON request bodies containing MongoDB operator keys ($ne, $gt, $lt, $regex, $where, $in).

Patch Information

At the time of NVD publication, the affected range spans FastGPT Community Edition 4.10.0 through 4.14.0. Monitor the FastGPT project repository for releases beyond 4.14.0 that address CVE-2026-79483, and upgrade to a fixed release once available. Verify the fix enforces type checking on JSON parameters and normalizes user input to primitive types before query construction.

Workarounds

Implement input coercion at the application boundary to cast expected chat history query parameters to strings before database calls.
Enforce authentication on /api/core/chat/getHistories at the ingress layer and drop unauthenticated requests.
Configure MongoDB with least-privilege service accounts so a successful injection cannot pivot to write or administrative operations.

bash

# Example WAF/nginx snippet: reject JSON bodies containing NoSQL operator keys
# Place inside the server or location block fronting FastGPT
location /api/core/chat/getHistories {
    if ($request_body ~* "\"\$(ne|gt|lt|gte|lte|in|nin|regex|where|expr)\"") {
        return 400;
    }
    proxy_pass http://fastgpt_upstream;
}