Joyland AI EUVD-2026-91025

The attacker must hold an adjacent network position (AV:A) shared with the victim and be able to intercept/tamper with content loaded into the app's WebView - implying interceptable or cleartext content delivery - in order to inject JavaScript; this is the core prerequisite (reflected by AT:P). No authentication is required (PR:N) and no user interaction is needed (UI:N). Baseline impact in the default state is limited to clipboard, app-internal storage, and arbitrary HTTP requests via the Weex 'stream' module. The escalated impact - entire filesystem, camera, microphone, and GPS - is CONDITIONAL on the installed app having previously been granted those OS-level permissions; on an install without those grants, exploitation is confined to the lower-impact set. The principal limiting factor is the adjacency/network-interception requirement, which prevents remote internet-based exploitation.

The CVSS 4.0 base score of 9.0 (Critical) reflects AV:A/AC:L/AT:P/PR:N/UI:N with full High impact to both vulnerable and subsequent systems (VC/VI/VA:H, SC/SI/SA:H), consistent with a WebView JavaScript-injection flaw (CWE-749, exposed dangerous method) that can escalate to filesystem, camera, microphone, and GPS access. However, several real-risk-dampening signals exist: the Adjacent attack vector (AV:A) means the attacker must share the victim's local network rather than attack over the internet, and AT:P signals a specific attack requirement (network interception position), so this is not internet-wide drive-by exploitation. No EPSS score, no CISA KEV listing, and no public POC are present in the provided data, so there is no public exploit identified at time of analysis and no evidence of active exploitation. The highest-impact outcomes (entire filesystem, camera, mic, GPS) are conditional on the app having been previously granted those OS permissions - a conflict worth highlighting: the headline 9.0 assumes the maximal permission-granted state, whereas the default no-prior-permission state is limited to clipboard, app-internal storage, and arbitrary HTTP via the Weex 'stream' module. Net: a genuinely serious client-side flaw for users on hostile Wi-Fi, but the adjacency requirement and consumer-app footprint make it a targeted/opportunistic local-network risk rather than a mass-exploitation emergency.

A victim uses the Joyland AI app while connected to an attacker-controlled or shared network (e.g., public Wi-Fi). The attacker intercepts and tampers with content loaded into the app's WebView, injecting JavaScript that abuses the Weex 'stream' module and exposed interfaces to read the clipboard and app-internal storage, and - if the app was previously granted device permissions - to access the filesystem, camera, microphone, and GPS. No public exploit code is identified in the provided data, and no user interaction (UI:N) is required once the attacker holds the network position.