Vulnerabilities
Splunk AI Toolkit 5.7.4 Vulnerability: Insecure Default Domain Allowlist Enables Low‑Privileged Data Exfiltration
A flaw in Splunk AI Toolkit versions below 5.7.4 allows low‑privileged users to trigger outbound HTTP requests to attacker‑controlled domains, potentially leaking data. The issue stems from an unrestricted default allowlist. Defenders should upgrade to 5.7.4 or configure explicit domain restrictions.
Splunk AI Toolkit versions prior to 5.7.4 contain a misconfiguration that permits low‑privileged users to direct the AI agent to send HTTP requests to arbitrary domains. This can lead to data exfiltration if an attacker controls the target server.
What Happened
- Insecure default allowlist in the AI Toolkit allows outbound requests to any domain.
- Users lacking admin or power roles can trigger these requests.
Impact
- Potential exfiltration of internal data over HTTP.
- CVSS v3.1 score 4.3 (Medium); CVE‑2026‑20265.
What to Do
- Upgrade to Splunk AI Toolkit 5.7.4 or later.
- If upgrade is not possible, edit local/mlspl.conf: set allowed_domains to a whitelist and ensure enforce_domain_validation=true.
- Alternatively, disable or remove the AI Toolkit app.
Detection
- No built‑in detection; monitor outbound HTTP traffic from the AI agent for anomalous destinations.