ai-fusion-video CVE-2026-110213

Exploitation requires a valid authenticated account on the ai-fusion-video application (PR:L - any low-privileged user suffices, no admin rights) and that the attacker can reach the image and video generation task /items endpoints over the network. The concrete prerequisite is that task records use predictable auto-increment ids and the /items endpoints perform no per-user ownership check, so the attacker supplies or enumerates arbitrary task ids belonging to other users. Limiting factors: the attacker must first hold an authenticated session, impact is confined to reading other users' generated images, videos, and output metadata (no modification or deletion), and only deployments running versions through 1.2.0 are affected.

This is a medium-severity information-disclosure flaw (CWE-639, IDOR) whose real-world risk is moderate and primarily confidentiality-bound. The provided CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:N/VC:L/VI:N/VA:N) is internally consistent with the description: network-reachable, low attack complexity, requires an authenticated low-privileged account (PR:L), no user interaction, and only low confidentiality impact with no integrity or availability impact. This is not a critical priority - the base score of 5.3 is driven down by the authentication requirement and the read-only nature of the exposure. No EPSS score was provided in the input, so exploitation probability cannot be quantified. The vulnerability is NOT listed in CISA KEV and no explicit POC=YES signal was provided, so there is no public exploit identified at time of analysis; however, a detailed technical write-up exists on hackmd and a VulnCheck advisory documents the issue, which lowers the bar for a capable attacker to reproduce it. The main signals align rather than conflict: a genuine but bounded authenticated IDOR against an obscure open-source product. The dominant limiting factor is product prevalence rather than exploit difficulty.

An attacker registers or obtains any low-privileged authenticated account on an ai-fusion-video (≤1.2.0) instance, then issues requests to the image and video generation task /items endpoints while substituting auto-increment task ids that belong to other users. Because the endpoints do not verify task ownership, the attacker retrieves other users' generated images, videos, and output metadata by simply iterating sequential ids. No public exploit code is identified at time of analysis, but a technical write-up describing the enumeration technique is publicly available.