Django CVE-2026-87890

Exploitable only when an application using django.contrib.gis passes attacker-controlled bytes values directly into a spatial lookup against a RasterField without wrapping them in a GDALRaster object; the malicious bytes must form a VRT document referencing an external raster source that GDAL will fetch. Limiting factors: the deployment must use GeoDjango raster fields and spatial lookups (a minority of Django apps), must forward raw untrusted bytes into those lookups, and the impact is confined to attacker-directed outbound network requests (SSRF) by the Django process rather than data compromise or code execution. Bytes representing valid hexadecimal geometries remain legitimately accepted, so generic byte input is not sufficient - the value must be interpretable as an external-referencing raster. This is not reachable in a default Django configuration that does not use GIS raster lookups.

This is a moderate-severity SSRF (CWE-918) that the vendor itself rates 5.3 and classifies as 'moderate' under Django's security policy. The provided CVSS 3.1 vector (AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N) reports network, low-complexity, unauthenticated access with only limited confidentiality impact and no integrity/availability impact - consistent with a server-side request forgery that coerces outbound network requests rather than code execution or data destruction. No EPSS score was supplied, this CVE is NOT on CISA KEV, and no public exploit code is referenced, so exploitation status is 'no public exploit identified at time of analysis.' The most important nuance is that the CVSS PR:N/AC:L rating overstates real-world reachability: per the vendor advisory the flaw is only exploitable by applications that pass attacker-controlled bytes directly into a GeoDjango spatial lookup against a RasterField - a narrow, non-default usage of django.contrib.gis. Treat this as a genuine but lower-priority patch-management item: raw CVSS looks like an easy unauthenticated network bug, but the practical precondition (GIS raster lookups fed untrusted bytes) sharply limits the affected population. This is a high-framework-prevalence but low-feature-prevalence issue, and it is an incomplete-fix regression of CVE-2026-15307, so organizations that patched the earlier CVE are still exposed.

A web application built on django.contrib.gis accepts user-supplied binary data and passes it as a bytes value into a spatial lookup against a RasterField. An attacker submits bytes containing a crafted VRT document that references an external (e.g. HTTP) raster source, and when Django prepares the lookup, GDAL opens the in-memory raster and issues a network request to the attacker-chosen URL as the Django process user. No public exploit code is identified at time of analysis.