ai-fusion-video EUVD-2026-99250

Requires a valid authenticated non-admin account on an ai-fusion-video instance (version ≤1.2.0); no admin privileges, user interaction, or special non-default configuration are needed because the ApiConfigController and SystemConfigController read endpoints lack authorization checks in the default code path. The attacker must be able to reach the application's HTTP API (network vector) and call the api-config get/list and system config endpoints. The primary limiting factor is the authentication requirement - an attacker needs legitimate low-privilege credentials or account access - which restricts exploitation to insiders, registered users, or anyone who has otherwise obtained a non-admin session, rather than fully anonymous internet attackers.

This is a genuine but bounded risk, best characterized as a credible second-order threat rather than a top-tier emergency. The CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:N/VC:H) is internally consistent with the description: network-reachable endpoints, low complexity, and a single confidentiality impact (VC:H, with VI:N/VA:N) reflecting pure secret disclosure via missing authorization (CWE-862). The key gating factor is PR:L - exploitation requires an already-authenticated, non-admin account, which meaningfully narrows the attacker population versus an unauthenticated flaw and argues against mass, internet-wide scanning. No EPSS score, KEV listing, or explicit POC flag was supplied in the input; however, a detailed public technical writeup (hackmd) and a VulnCheck advisory exist, which lowers the bar for a motivated authenticated attacker even without packaged exploit code. The real-world danger is amplified by what is exposed: harvesting AI-provider API keys, proxy passwords, and mail_password enables downstream billing abuse and authenticated outbound mail spoofing as the platform, so the blast radius extends beyond this one app to the attacker-controlled abuse of third-party services and the platform's reputation. Net: prioritize for any operator actually running this product with multi-user accounts, but it is not a drop-everything network RCE.

A low-privileged but legitimately registered user of an ai-fusion-video deployment sends authenticated requests to the api-config get/list and system config read endpoints, which fail to enforce admin authorization. The response returns stored secrets - AI provider API keys, proxy passwords, and the mail_password - which the attacker then uses to run up charges on the platform's AI/API accounts and to send mail that appears to originate from the platform. A detailed technical writeup describing the broken access control is publicly available, though no packaged exploit code was confirmed in the provided data.