Executive brief
Emlog Pro's AI service module unconditionally disables TLS certificate validation when communicating with external AI providers (like ChatGPT or image generation services). An attacker positioned on the network path can intercept these connections, steal the API key used to authenticate with the AI service, and inject fake AI responses that the application will execute—potentially leading to unauthorized database queries or configuration changes.
Technical details
The vulnerability is improper certificate validation (CWE-295) in the AI service HTTP client. Four functions in include/service/ai.php (sendStream, sendImageRequest, send, and fetchSearchHtml) issue HTTPS requests to configured LLM providers with CURLOPT_SSL_VERIFYPEER and CURLOPT_SSL_VERIFYHOST unconditionally disabled; there is no configuration option to re-enable verification. An attacker with network adjacency (e.g., shared network segment, compromised router, or ARP spoofing) can perform a man-in-the-middle attack by presenting a self-signed certificate. Because SSL verification is disabled, the application accepts the attacker's certificate, allowing the attacker to read the Authorization Bearer token from outbound requests in plaintext and optionally inject crafted responses that trigger tool-call handlers (query_database, update_config). No patch is currently available.
Affected products
- Emlog Emlog Pro through 2.6.23
Timeline
- 2026-08-01: disclosed: GitHub security advisory GHSA-hf85-99vj-m4c5 published
- 2026-08-03: advisory: NVD entry CVE-2026-67598 published