web.py EUVD-2026-85403

Exploitation requires a specific deployment: the web.py 0.76 application must precompile templates from a directory that the attacker can write to, and must later render those templates through the precompiled template loader - that combination is what bypasses the built-in template sandbox. Concrete prerequisites/limiters: (1) attacker must first obtain write access to the template source directory (e.g. via an upload feature, misconfigured permissions, or existing low-privilege foothold - reflected as PR:L); (2) the application must use the precompiled-template code path rather than direct sandboxed rendering; (3) exploitation is not automatic against a default install (reflected as AC:H). It is NOT a remote-unauthenticated exploit against stock configurations; the writable-template-directory and precompiled-loader conditions are the decisive gating factors defenders should confirm before prioritizing.

This is a genuine RCE-class flaw (CWE-94, code injection leading to arbitrary Python and OS command execution) but its real-world risk is materially reduced by its preconditions rather than being a broadly sprayable network exploit. The vendor CVSS vector (CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:C/C:H/I:H/A:H, 8.5 HIGH) is internally consistent with the description: AC:H and PR:L reflect that exploitation is not turnkey - the attacker must already have some access AND the application must be deployed in a specific way (precompiling templates from a directory the attacker can write to, then rendering via the precompiled loader). No EPSS score was supplied, the CVE is NOT on CISA KEV, and the only exploitation-related reference is a TraceLoom write-up that reads as vulnerability analysis rather than a packaged, weaponized exploit - so exploitation status is: no public exploit identified at time of analysis. The S:C (scope change) claim is defensible (escaping the template sandbox to run host-level code) but arguably should be S:U, as ordinary host RCE within a single security authority; this is the one metric worth verifying with the reporter. Net: a real, patch-worthy issue for anyone running web.py 0.76 with precompiled templates from writable directories, but not a mass-exploitation priority given the conditional deployment requirement, authenticated/write prerequisite, and absence of KEV/EPSS/confirmed POC signals.

An attacker who can write files into a directory that a web.py 0.76 application uses for template precompilation drops a malicious template containing code the built-in sandbox is meant to reject. When the application later loads and renders that template through the precompiled template loader, the security checks are bypassed and the attacker's Python executes on the server, yielding OS command execution. A public exploit is not confirmed - only a third-party analysis write-up (TraceLoom) exists - and the attack chain hinges on the attacker first obtaining write access to the template directory.