sharp EUVD-2026-106916

The application must use sharp earlier than 0.35.0 (with the bundled vulnerable libvips, or a system libvips older than 8.18.3) AND must pass attacker-controlled images into sharp for decoding. Exploitation specifically requires the crafted file to reach the GIF, TIFF or VIPS loaders (VipsForeignLoadNsgif, VipsForeignLoadTiff, VipsForeignLoadVips); other formats are unaffected. Limiting factors: only these three decoders are in scope, so the documented workaround sharp.block({ operation: ['VipsForeignLoadNsgif','VipsForeignLoadTiff','VipsForeignLoadVips'] }) fully mitigates it, and applications that never process untrusted images or that already restrict accepted formats are not exposed. No victim user interaction is needed because processing is automated server-side; whether authentication is required to submit an image is deployment-dependent and not fixed by the CVE.

This is a CWE-1395 umbrella advisory: sharp < 0.35.0 bundles a libvips build affected by five distinct image-decoder flaws (CVE-2026-33327, -33328, -35590, -35591, -69242) in the GIF, TIFF and VIPS loaders, two rated High under CVSSv4. The provided CVSS:4.0 vector (base 8.4) specifies AV:L/PR:L, implying local access and some privilege, yet the description explicitly frames the attack as remote delivery of a crafted image to 'applications using sharp to process untrusted input' - the classic server-side image-upload scenario - so attack vector should likely be NETWORK, not LOCAL; verify positioning against your deployment. The impact metrics (VC:L, VI:H, VA:H) point to memory-corruption outcomes (integrity and availability loss, limited disclosure) rather than the input 'Information Disclosure' tag, which understates the integrity/availability risk. There is no CISA KEV listing and no public exploit identified, and no EPSS score was supplied, so there is no evidence of active or imminent widespread exploitation - this reads as a high-severity-but-not-yet-exploited patch-hygiene priority. The real driver of risk is sharp's enormous deployment footprint plus the ease of reaching the vulnerable decoders whenever untrusted images are processed; treat as a prompt patch where untrusted image input is accepted, but not an emergency absent POC/KEV signals.

An attacker uploads a specially crafted GIF, TIFF or VIPS file to a web or backend service that uses sharp < 0.35.0 to resize, convert or inspect user-supplied images. When sharp hands the file to the bundled libvips GIF/TIFF/VIPS decoder, the malformed structures trigger one of the underlying libvips flaws, causing memory corruption that can crash the process (DoS) or potentially corrupt memory for further impact. No public exploit code has been identified for this specific advisory, and exploitation depends on the application actually decoding the affected formats.