CVE-2026-76393: Splunk AI Toolkit Race Condition Flaw

CVE-2026-76393 Overview

CVE-2026-76393 is a race condition vulnerability in Splunk AI Toolkit versions below 6.0.0. An authenticated user with model upload permissions can overwrite a model being uploaded by another user by sending a concurrent upload request for the same model name. The result is a model lookup entry that references attacker-controlled content. The flaw stems from the toolkit failing to verify that uploaded content belongs to the request that created the lookup entry, classified as a concurrent execution using shared resource with improper synchronization [CWE-362].

Critical Impact
A low-privileged user can substitute a legitimate machine learning model with attacker-controlled content, tainting downstream inference and analytics that depend on the compromised model lookup.

Affected Products

Splunk AI Toolkit versions below 6.0.0
Deployments integrating the Splunk Machine Learning Toolkit for model uploads
Splunk environments where multiple users share model upload permissions

Discovery Timeline

2026-08-19 - CVE-2026-76393 published to NVD
2026-08-20 - Last updated in NVD database

Technical Details for CVE-2026-76393

Vulnerability Analysis

The vulnerability is a Time-of-Check to Time-of-Use (TOCTOU) race condition in the model upload workflow of Splunk AI Toolkit. When a user uploads a model, the toolkit creates a model lookup entry that points to the uploaded content. The toolkit does not bind the lookup entry to the specific upload request that created it. An attacker who can also upload models can race a legitimate upload by submitting a concurrent request using the same model name. The lookup entry then references the attacker's uploaded content rather than the original user's data.

Because the compromised entity is a machine learning model consumed by downstream searches, dashboards, and detections, the integrity impact extends beyond the initial file substitution. Any analytics, alerts, or automated actions that rely on the tainted model may produce attacker-influenced results.

Root Cause

The root cause is missing synchronization between the upload transaction and the model lookup registration. The toolkit does not verify that the content persisted on disk originated from the request that registered the lookup entry, allowing concurrent writes to the same model name to interleave.

Attack Vector

Exploitation requires network access, low privileges (model upload capability), and user interaction from the victim uploader. The attacker must time a concurrent upload request against a legitimate upload for the same model name. Attack complexity is high because the attacker must win the race window. Successful exploitation yields high integrity impact against the model artifact.

No public proof-of-concept exploit is available, and the vulnerability is not listed in the CISA Known Exploited Vulnerabilities catalog. Refer to the Splunk Security Advisory SVD-2026-0808 for vendor technical details.

Detection Methods for CVE-2026-76393

Indicators of Compromise

Multiple concurrent upload requests targeting the same model name from different user sessions within a short time window
Model lookup entries whose file content hash does not match the audit log of the user who registered the entry
Unexpected changes in model file size, checksum, or last-modified timestamp shortly after a legitimate upload event

Detection Strategies

Audit Splunk internal logs (_internal index) for POST requests to the Splunk AI Toolkit model upload endpoints correlated by model name and timestamp
Alert when two or more distinct users submit uploads for the same model name within the same second
Baseline expected model uploaders per model and flag deviations where an uploader is not the model's typical owner

Monitoring Recommendations

Continuously monitor Splunk role assignments granting model upload capability and review membership regularly
Track integrity of model lookup files by comparing periodic checksums against a known-good manifest
Review Splunk AI Toolkit search job history for unexpected model references or lookup replacements

How to Mitigate CVE-2026-76393

Immediate Actions Required

Upgrade Splunk AI Toolkit to version 6.0.0 or later, which addresses the race condition
Restrict the Splunk role capability to upload models to a minimal set of trusted users
Review existing model lookup entries for unexpected content and re-upload trusted models where necessary

Patch Information

Splunk has released Splunk AI Toolkit 6.0.0 to remediate CVE-2026-76393. Consult the Splunk Security Advisory SVD-2026-0808 for exact patched versions and upgrade guidance. Apply the upgrade in accordance with the standard Splunk app update process for your deployment topology.

Workarounds

Limit model upload permissions to a small group of administrators until the upgrade is applied
Enforce naming conventions that scope model names to individual users or teams to reduce collision opportunities
Serialize model upload operations through organizational process controls until patched

bash

# Configuration example: restrict model upload capability in Splunk authorize.conf
# Remove the upload_lookup_files capability from non-admin roles that use the AI Toolkit
[role_mltk_user]
upload_lookup_files = disabled

# Verify installed Splunk AI Toolkit version
$SPLUNK_HOME/bin/splunk display app Splunk_ML_Toolkit