### Summary
When the LaTeX backend is configured to render TikZ pictures with the Tectonic engine (`LatexBackendOptions(tikz_engine="tectonic")`), the TikZ body and the document preamble from the input file are compiled by Tectonic without any restriction on TeX's file primitives. A crafted `.tex` file can read files the process can read and write files at paths the process can write. If the caller also sets `tikz_engine_allow_shell_escape=True`, the document can run shell commands through `\write18`.
### Details
`docling/backend/latex/engines/tectonic.py` writes the picture source into a temporary document and runs the `tectonic` binary on it. `_resolve_local_dependency` only decides which files are staged into the temporary directory. It does not limit what TeX opens at compile time, so:
- `\input{/abs/path}`, `\verbatiminput{...}` and `\openin` read arbitrary files, and their content can be typeset into the rendered picture that ends up in the conversion output. For an in-memory `DocumentStream`, staging is skipped entirely.
- `\immediate\openout` / `\write` create or overwrite files outside the temporary directory. Neither needs shell escape. Tectonic's `--untrusted` flag does not block them either.
Separately, the `TectonicEngine` constructor defaults to `allow_shell_escape=True`. The LaTeX backend always passes `tikz_engine_allow_shell_escape` explicitly (default `False`), so this default only affects code that instantiates `TectonicEngine` directly.
### Affected configurations
- Affected: callers that set `tikz_engine="tectonic"`, have a `tectonic` binary available (on `PATH` or in the docling cache), and convert untrusted LaTeX.
- Not affected: the default configuration (`tikz_engine=None`). docling does not install `tectonic` automatically, and the CLI exposes no option to enable it.
### Impact
Disclosure of local files readable by the process, and creation or overwrite of files writable by the process. Writing to startup or configuration files can lead to code execution. With shell escape enabled, commands run directly.
### Proof of concept
```latex
\documentclass{article}
\usepackage{tikz}
\begin{document}
\begin{tikzpicture}
\immediate\openout15=/tmp/docling-tectonic-poc.txt
\immediate\write15{written by the input document}
\immediate\closeout15
\node {x};
\end{tikzpicture}
\end{document}
```
Converting this file with `LatexBackendOptions(tikz_engine="tectonic")` creates `/tmp/docling-tectonic-poc.txt`.
### Patches
Mitigated in docling 2.132.0 by [#4419](https://github.com/docling-project/docling/pull/4419):
- `TectonicEngine` now defaults to `allow_shell_escape=False`, and without shell escape Tectonic runs with `--untrusted` and `--only-cached`.
- Before compiling, docling checks the generated document and every staged file. It skips rendering, and keeps the diagram as TikZ code, when the source uses the `\openin`, `\openout`, `\XeTeXpicfile` or `\XeTeXpdffile` primitives, or names an absolute, home, drive-letter or parent-directory path in `\input`, `\include`, `\includegraphics`, `\InputIfFileExists` or `\graphicspath`.
This check is best-effort. It reads the source text, and TeX can open files in ways a text check cannot see, for example through package commands or names built by macros. As documented in the [pipeline options reference](https://github.com/docling-project/docling/blob/main/docs/reference/pipeline_options.md), untrusted LaTeX rendered with `tikz_engine="tectonic"` should still be processed in an isolated environment.
### Workarounds
Upgrade to 2.132.0. For older versions:
- Do not enable `tikz_engine="tectonic"` for untrusted input, or run the conversion in an OS-level sandbox (container, no host mounts, read-only filesystem, no network).
- Never set `tikz_engine_allow_shell_escape=True` for untrusted input.
MCP Atlassian versions prior to 0.22.0 contain a path traversal vulnerability in the upload_attachment functionality. This flaw allows an attacker with limited privileges to specify an absolute or traversal file path during attachment uploads, enabling arbitrary local file reads. The issue stems from an incomplete fix for CVE-2026-27825 that only protected download destinations but did not restrict source paths for uploads. The vulnerability is fixed in version 0.22.0.
MCP Atlassian versions prior to 0.22.0 store OAuth access and refresh tokens in plaintext fallback files with group-readable permissions. This allows local users or processes with group access to read and reuse these tokens, potentially gaining unauthorized access to Atlassian resources. The issue is fixed in version 0.22.0.
A cross-site scripting (XSS) vulnerability exists in TryGhost Ghost versions 6.22.1 through 6.64.0. The issue arises because the default local storage adapter does not enforce content type restrictions on uploaded files, allowing scripts to be hosted on the site's domain. This can lead to compromise of other staff users' admin sessions. The vulnerability is fixed in version 6.64.0.
CVE-2026-58880 is a high-severity elevation of privilege vulnerability in Google Android affecting versions 16, 16-qpr2, and 17. It arises from a race condition in the handle_app_val_response function of the btif_rc.cc component. Exploitation does not require user interaction and can lead to local code execution with elevated privileges.
CVE-2026-55266 is a high-severity vulnerability in Google Android affecting versions 14, 15, 16, 16-qpr2, and 17. It involves a possible out-of-bounds write in the qsort function of libufdt_sysdeps_vendor.c due to resource exhaustion. This flaw can lead to a denial of service and local privilege escalation without requiring user interaction.
Plane is an open-source project management tool. Prior to 1.4.0, Plane trusts email addresses returned by Gitea OAuth and by self-managed GitLab OAuth deployments where email confirmation is disabled, without verifying that the provider authenticated ownership of the address. An attacker can set an OAuth identity's unverified provider email to a victim's address, which Plane matches directly to the victim's existing local account. The attacker can then log in to the victim's Plane account without knowing the victim's password. GitHub, GitLab.com, and Google are not affected because those providers return verified email addresses. This issue is fixed in 1.4.0.
A time-of-check to time-of-use (TOCTOU) race condition exists in Progress Telerik Fiddler Classic for Windows versions prior to 6.0.20262.10021. This vulnerability occurs during the installation of the HTTPS interception root certificate, where a temporary file is written in a user-writable location and then imported by an elevated helper application. A local low-privilege attacker can replace this temporary file before it is read by the elevated process, causing an attacker-controlled root certificate to be installed in the Local Computer Trusted Root Certification Authorities store. Exploitation requires user initiation and approval of an elevation prompt.
CVE-2026-63269 is a server-side request forgery (SSRF) vulnerability in LibreOffice version 26.2. The issue arises when LibreOffice links to audio and video files, specifically on Linux where it uses GStreamer to play media. An attacker could craft an HLS playlist that causes GStreamer to read local and remote files listed in the playlist during document loading, potentially exposing their contents. The vulnerability has been addressed in fixed versions by preventing LibreOffice from following playlists that reference further resources and by placing linked media under link update control.
CVE-2026-63267 is a server-side request forgery (SSRF) vulnerability in LibreOffice Calc version 26.2. It allows a document to link a cell range to an external CSV data source, which is fetched when the document loads. This behavior could enable reading local files into the spreadsheet or making requests to arbitrary hosts chosen by the document. The vulnerability is addressed in fixed versions where external data links are updated under the same control as other spreadsheet links.
LibreOffice Calc can link a cell range to an external data source, and the link is saved in the document. A link of the sql type could name a folder of local text files as a database, so opening a document could read a local text file into the sheet. In fixed versions only the csv, html and xml data providers are restored when a document is loaded.
CVE-2026-105249 is a use-after-free vulnerability in the vgmstream software up to version r2117. It affects the make_group_random function in the src/meta/txtp_process.c file, specifically within the TXTP File Handler component. The vulnerability requires local access to exploit and has a low severity score. A patch identified by commit ae37662ad626254ddd96ad69ac263792d7a92024 is recommended to fix this issue.
Feehi CMS version 2.1.1 and earlier contain a Server-Side Request Forgery (SSRF) vulnerability in the UEditor catchimage endpoint. The vulnerability arises because the private-IP validation does not block loopback or link-local addresses, enabling an attacker to make the server send HTTP requests to internal services. This could lead to unauthorized information disclosure from internal network resources.
CVE-2026-78862 is a vulnerability in the Mercusys AC12 V2 router that allows a local attacker to execute arbitrary code via the UART serial interface on the device's printed circuit board. This vulnerability has a CVSS score of 6.8, indicating a medium severity level. No affected versions or patches are explicitly stated in the available data.
CVE-2026-78861 is a high-severity vulnerability in Mercusys AC12 V2 that allows a local attacker to execute arbitrary code by leveraging a hardcoded 512-bit RSA private key. This flaw enables unauthorized code execution without requiring privileges or user interaction. The vulnerability has a CVSS score of 7.7, indicating significant confidentiality and integrity impact but no availability impact. No affected versions or patch information are provided, and no known exploits are reported in the wild.
CVE-2026-78860 is a high-severity vulnerability in Mercusys AC12 V2 that allows a local attacker to execute arbitrary code by exploiting the storage of information in plaintext. This vulnerability has a CVSS score of 7.8, indicating significant impact on confidentiality, integrity, and availability. No affected versions or patch information are provided, and no known exploits are reported in the wild.
The MPG WordPress plugin before 4.2.3 does not validate that the dataset source supplied when importing a project is a remote URL before treating it as a local filesystem path and copying that file into a publicly accessible uploads folder. This makes it possible for users with the Editor role and above to read the contents of arbitrary files on the server, with the copied file then retrievable by unauthenticated visitors.
The WPZOOM Connect: AI Chat, Click to Chat, Social Icons & Share Buttons plugin for WordPress is vulnerable to Sensitive Information Exposure in all versions up to, and including, 4.7.3 via the 'x-yamidoo-signature (attacker-obtained via inline_js identify payload)' parameter. This makes it possible for unauthenticated attackers to extract the full customer card — including name, WordPress user ID, order history, order totals, purchased products, payment method labels, and EDD Software Licensing license keys with status and activation counts — for any arbitrary victim email address on the site. Exploitation requires the attacker to register a WooCommerce customer or subscriber-level account with a crafted email address whose local part encodes the target timestamp and victim email, allowing the signature printed into the page HTML by inline_js() to pass verify_request() for an arbitrary victim; both the share_customer_data and identify_logged_in settings are enabled by default, so no non-default configuration is required.