Emlog Pro EUVD-2026-52405

Exploitation requires the AI/LLM integration in include/service/ai.php to be configured and actively making outbound HTTPS requests, and it requires the attacker to occupy a man-in-the-middle position on the network path between the Emlog Pro server and the configured LLM provider - the description explicitly scopes this to 'network-adjacent' attackers, not arbitrary remote hosts. No application credentials are needed (PR:N) and no user interaction is required (UI:N), but the on-path prerequisite (reflected as AC:H) is the primary limiting factor: without the ability to intercept and answer the TLS connection, the disabled verification cannot be leveraged. Deployments whose AI traffic never leaves a trusted, controlled network path, or that do not use the AI feature at all, are not realistically exploitable.

The vendor CVSS 4.0 vector (AV:N/AC:H/AT:N/PR:N/UI:N/VC:H/VI:H/VA:N) scores 9.1 Critical, driven by high confidentiality (API-key theft) and high integrity (response injection) with no availability impact. AC:H is the key limiter: exploitation is not push-button because the attacker must first achieve a man-in-the-middle position on the path between the Emlog server and its LLM provider, which the description scopes as 'network-adjacent' rather than any arbitrary remote host - so the practical attacker population is far smaller than a typical AV:N/AC:L flaw. No EPSS, SSVC, POC, or KEV signals were provided, so all exploitation indicators point to theoretical-but-serious rather than mass-exploited. Note the tension between the official AV:N and the 'network-adjacent' wording: the true vector is best modeled as adjacent/on-path, and defenders should not read AV:N as 'exploitable by any internet host.' Treat this as high priority where AI traffic crosses untrusted or shared networks and lower where egress is tightly controlled.

An attacker positioned on the network path between an Emlog Pro server and its LLM provider - on a shared hosting segment, a compromised gateway, or an untrusted egress link - presents a forged TLS certificate that the vulnerable cURL client accepts without complaint. The attacker records the Authorization Bearer API key from the intercepted request, then returns a crafted 'AI response' designed to drive the update_config or query_database tool handlers, converting stolen credentials and response injection into deeper compromise. No public exploit code is identified at time of analysis, and success requires the attacker to already hold an on-path position (CVSS AC:H).