CVE-2026-63734: Surrealdb Surrealdb DOS Vulnerability

CVE-2026-63734 Overview

CVE-2026-63734 is a denial of service vulnerability in SurrealDB versions before 3.2.0. The flaw resides in the SurrealML header parser exposed through the /ml/import endpoint. An authenticated user holding the Owner role can upload a malformed .surml file that triggers unchecked unwrap calls in the parser. The resulting panic aborts the entire SurrealDB server process, denying service to every database hosted on that instance. The vulnerability is classified under [CWE-20] Improper Input Validation and affects the SurrealML machine learning model import path.

Critical Impact
An authenticated Owner-role attacker can crash the SurrealDB server process with a single malformed .surml upload, taking all hosted databases offline.

Affected Products

SurrealDB versions prior to 3.2.0
SurrealDB deployments exposing the /ml/import endpoint
Multi-tenant SurrealDB instances hosting multiple databases on a single process

Discovery Timeline

2026-07-20 - CVE-2026-63734 published to NVD
2026-07-22 - Last updated in NVD database

Technical Details for CVE-2026-63734

Vulnerability Analysis

SurrealDB integrates SurrealML to import serialized machine learning models via the HTTP /ml/import endpoint. The endpoint accepts .surml files whose headers describe model metadata, including numeric fields such as input dimensions. The header parser calls unwrap on the results of type conversions and field lookups without validating the underlying data. When a supplied field is non-numeric or otherwise malformed, the unwrap returns None or Err, and the Rust runtime raises a panic. Because the parser runs inside the main server process, the panic propagates and terminates the entire SurrealDB instance, denying service to all tenants and databases hosted on it.

Root Cause

The root cause is improper input validation combined with unsafe error handling. The SurrealML header parser trusts client-supplied bytes and uses unwrap where fallible parsing operations should return recoverable errors. There is no bounded validation of numeric fields such as input dimensions before conversion, and there is no panic isolation around the ML import task.

Attack Vector

Exploitation requires network access to the SurrealDB HTTP interface and valid credentials with the Owner role. An attacker sends a crafted .surml file to /ml/import containing a non-numeric value in a header field expected to be an integer, or another malformed structural element. The parser panics, the process exits, and every database served by that node becomes unavailable until an operator restarts the service. Because the impact affects availability only, confidentiality and integrity are not directly at risk. See the VulnCheck Advisory on SurrealDB for additional technical detail.

Detection Methods for CVE-2026-63734

Indicators of Compromise

Unexpected POST requests to the /ml/import endpoint followed by SurrealDB process termination or restart events.
Server logs showing a Rust panic message referencing the SurrealML header parser or an unwrap on a None/Err value.
Repeated .surml upload attempts from a single Owner-role account within a short window.
Process supervisor entries recording abnormal SurrealDB exits correlated with ML import activity.

Detection Strategies

Enable request logging on the SurrealDB HTTP listener and alert on any traffic hitting /ml/import from accounts or source IPs not authorized to publish models.
Correlate SurrealDB process exit codes with the last inbound HTTP request to identify import-triggered crashes.
Monitor authentication logs for use of Owner-role credentials outside expected administrative hosts.

Monitoring Recommendations

Ingest SurrealDB stdout, stderr, and audit logs into a central logging pipeline and search for panic strings originating from the ML module.
Track uptime and restart counts of each SurrealDB node and alert on unexpected restarts.
Baseline the volume and size of .surml uploads and alert on deviations.

How to Mitigate CVE-2026-63734

Immediate Actions Required

Upgrade SurrealDB to version 3.2.0 or later on all nodes.
Rotate Owner-role credentials and verify that Owner privileges are limited to trusted operators.
Restrict network exposure of the /ml/import endpoint to trusted management networks using a reverse proxy or firewall rule.
Audit historical logs for prior .surml upload attempts and unexplained SurrealDB restarts.

Patch Information

SurrealDB has addressed CVE-2026-63734 in release 3.2.0. The fix replaces unchecked unwrap calls in the SurrealML header parser with proper error handling so malformed uploads return an error response rather than aborting the process. Refer to the GitHub Security Advisory GHSA-jwr6-6444-28xv for full release notes.

Workarounds

Disable the /ml/import endpoint at the reverse proxy layer if machine learning model import is not required.
Reduce the number of accounts holding the Owner role to the minimum necessary.
Run SurrealDB under a process supervisor such as systemd with automatic restart to shorten outage duration until patching is complete.
Deploy per-tenant SurrealDB instances rather than multi-tenant nodes to limit blast radius of a single crash.

bash

# Verify installed SurrealDB version and upgrade
surreal version
# Upgrade to a patched release (3.2.0 or later)
curl -sSf https://install.surrealdb.com | sh
# Restrict /ml/import at an nginx reverse proxy
# location = /ml/import {
#     allow 10.0.0.0/24;   # management network
#     deny  all;
#     proxy_pass http://surrealdb_upstream;
# }