AI Engine CVE-2026-75798

No special conditions are required beyond network access to a WordPress site running AI Engine 3.4.0 through 3.7.1 with the plugin active. The enabling condition is the plugin's own token - distributed to all anonymous visitors - which serves as the sole authorization gate for an administration-only feature; any attacker who can reach the site can obtain and use this token without credentials. No non-default configuration, user interaction, or elevated privilege is needed. Exploitation is bounded to sites with AI Engine installed and active; impact scales with the site owner's AI provider account limits and the attacker's query volume.

The CVSS 3.1 score of 5.3 Medium is consistent with the exploitation profile: network-reachable, no complexity, no authentication, no user interaction, but impact constrained to I:L with no confidentiality or availability consequence. The EPSS score of 0.17% (6th percentile) is low, indicating that automated exploitation at scale has not been widely observed - yet the existence of a public proof-of-concept from WPScan meaningfully lowers the barrier for opportunistic abuse. The real-world risk is primarily financial: attackers can exhaust AI API credits, trigger overage fees, or saturate provider rate limits, degrading service for legitimate users. This is not a data-exfiltration or system-compromise scenario. The absence of a CISA KEV entry corroborates that broad active exploitation has not been confirmed. Overall, the vulnerability warrants prompt patching given the trivial exploitation path and financial abuse potential, but does not represent an urgent incident-response scenario in the absence of KEV confirmation.

An unauthenticated attacker discovers a WordPress site running AI Engine before 3.7.2 - via passive scanning or targeted enumeration - and uses the token the plugin distributes to anonymous visitors to call the unprotected administration-only AI endpoint. The attacker submits a high volume of AI queries (e.g., bulk text generation, prompt probing), which are forwarded to the site owner's AI provider account, consuming API credits and potentially triggering overage charges or rate-limit errors for legitimate users. A public proof-of-concept lowering the technical barrier for this attack is available via WPScan.