Google Android CVE-2026-58944

Requires a locally installed, untrusted application running on the target device - the CVSS AV:L vector means there is no remote network path; an attacker must first get code onto the device (e.g., via a malicious app). Given PR:N/UI:N, that app needs no special permissions and no user interaction to trigger the confused deputy in moveTaskToFrontLocked and perform the background activity launch. Exploitation is limited to Android 14, 15, 16, 16-qpr2, and 17; the primary limiting factor is the local-only attack surface (an app must already be present/executing), and SSVC marks the flaw as not automatable, constraining mass exploitation.

This is a genuine but not emergency-level priority. The CVSS vector (AV:L/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H, 8.4 HIGH) describes a local attack: the attacker needs code already running on the device (a malicious or compromised app), not network reach, so AV:L is the dominant risk-limiting factor despite PR:N. The SSVC decision framework corroborates a measured posture - Exploitation: none (no observed or public exploitation), Automatable: no, Technical Impact: total - which maps to the vuln.today 'this_week' priority rather than act_now/emergency. There is no CISA KEV listing and no EPSS score supplied, so widespread or active exploitation is not indicated; combined with SSVC Exploitation:none, the real-world near-term risk is lower than the 8.4 base score alone would suggest. The high score is driven by the 'total' technical impact (full C/I/A compromise once triggered) rather than by ease of remote reach. One point to verify: SSVC marks Automatable:no, consistent with the confused-deputy BAL nature requiring a locally installed app, which argues against mass/worm-style exploitation.

A user installs a malicious or trojanized app that requests no special permissions. The app abuses the confused deputy in moveTaskToFrontLocked to perform a background activity launch, bypassing BAL restrictions to bring its own (or a spoofed) activity to the foreground without user interaction, escalating its effective privilege locally. No public exploit code is identified at time of analysis, and the precise trigger sequence is not described in the available sources.