Amazon Web Services has patched a critical vulnerability in Amazon Neptune, its managed graph database, that carries a CVSS score of 9.9. The flaw, tracked as CVE-2026-110812, is an unsafe deserialization issue in Apache TinkerPop Gremlin and was described in AWS security bulletin 2026-136-AWS, published October 15.
CVE-2026-110812: Unsafe Deserialization in the Graph-Import Path
According to the bulletin, the vulnerability lies in the graph-import path functionality of Apache TinkerPop Gremlin, the graph traversal framework that Neptune’s query engine uses. The weakness is the deserialization of untrusted data. Amazon Neptune engine versions 1.2.0.0 and later, before the patched releases, are affected.
An Authenticated Principal With Query Permissions Can Trigger It
Exploitation requires an authenticated principal that holds Neptune data-plane permissions, which are the query permissions. Such a principal can send a crafted request that makes the query engine process untrusted input unsafely. The result is compromise of the query engine environment.
The authentication requirement narrows who can attack, but the permission involved is the ordinary one that applications and users need to run queries, so any holder of query credentials is a potential source.
Scope of the Affected Version Range
The affected range is open-ended at the low end: engine version 1.2.0.0 and every later release before the patched builds. The fixed builds sit on three release lines, 1.4.7.1, 1.4.8.1 and 1.5.0.0, so customers on each line have a corresponding target. The advisory is numbered 2026-136-AWS, and it classifies the weakness as the deserialization of untrusted data in the graph-import path of Apache TinkerPop Gremlin.
No Proof-of-Concept and No Known Exploitation
AWS reports no public proof-of-concept and no confirmed exploitation of CVE-2026-110812. The 9.9 score reflects the severity of the outcome rather than any evidence of attacks, and AWS has not described any customer impact in the bulletin.
Fixed Engine Versions and Upgrade Behavior
AWS has published the fix as new Neptune engine releases, and the way customers receive them varies by how their clusters are configured.
The flaw is fixed in Neptune engine versions 1.4.7.1, 1.4.8.1 and 1.5.0.0. How a customer receives the update depends on cluster settings.
Clusters With Auto Minor Version Upgrade Get the Fix in the Next Maintenance Window
Clusters configured with auto minor version upgrade receive the update during their next maintenance window. Clusters without the setting need a manual upgrade, so those customers carry the responsibility for moving to one of the three patched versions.
Restricting neptune-db Query Permissions Until the Upgrade
Until a cluster is upgraded, AWS advises granting neptune-db query permissions only to trusted principals. Because the attack depends on query access, limiting who holds it is the stopgap the bulletin offers while the patched version is applied.
What the Rating Means for Enterprise Graph Databases
Enterprises use Neptune graph databases to store connected data, and a compromise of the query engine environment gives an attacker a foothold inside that service. The takeover risk applies to anyone holding query credentials, which makes the management of IAM permissions for neptune-db the practical front line while patching is incomplete.
The vulnerable code is in an open-source component, Apache TinkerPop Gremlin, but the bulletin presents the fix as Neptune engine releases rather than as a change customers make themselves to the component. That distinction means customers cannot remediate by patching the library independently and must move their clusters to the listed engine versions.
The bulletin leaves one practical gap for customers: it does not say whether any clusters have already been probed, and AWS reports no confirmed exploitation. Customers who run Neptune on engine versions earlier than the patched releases remain within the affected range until their clusters are upgraded, whether through a maintenance window or manually.
