CVE-2026-47214: Docling HTML Backend URI/Path Handling Vulnerability
Docling, a document processing library that converts various file formats and integrates with AI systems, contains a vulnerability in its HTML parser that mishandles URIs and file paths. An attacker can exploit this by crafting a malicious document that, when processed by a user, could leak sensitive information or cause the application to become unavailable. The vulnerability affects versions prior to 2.94.0 and requires user interaction to trigger.
Source data · NVD / CISA · public domain
- CVSS
- 3.1 · 7.1 HIGH · CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:L
- Weaknesses (CWE)
- CWE-400, CWE-73
- Affected products
- 1 configuration(s)
- Published / Modified
- 2026-06-26 / 2026-07-02
NVD description (verbatim)
Docling simplifies document processing by parsing diverse formats and providing integrations with the generative AI ecosystem. Prior to 2.94.0, the HTML backend has unsafe URI and path handling. This vulnerability is fixed in 2.94.0.
2 reference(s) · View on NVD →
SEC.co analysis · AI-assisted, reviewed against source
Technical summary
CVE-2026-47214 exploits unsafe URI and path handling in Docling's HTML backend. The vulnerability maps to CWE-400 (Uncontrolled Resource Consumption) and CWE-73 (External Control of File Name or Path), indicating that the HTML parser fails to properly validate and sanitize resource references before processing. This allows attackers to craft documents containing malicious URI references that trigger uncontrolled resource consumption or path traversal conditions. The CVSS 3.1 vector (AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:L) reflects that exploitation requires network access and user interaction, but results in high confidentiality impact and limited availability impact.
Business impact
Organizations using Docling for document processing—particularly those integrating it into AI pipelines or customer-facing document analysis workflows—face two primary risks: confidential information disclosure when processing untrusted documents, and service degradation if an attacker sends documents that exhaust computational resources. The user-interaction requirement limits automated attack scenarios but remains significant if your application processes documents from external sources or allows users to upload files. Downstream AI applications relying on Docling output could also be exposed to manipulated or incomplete data.
Affected systems
All versions of Docling prior to 2.94.0 are affected. The vulnerability is specific to the HTML backend's URI and path handling, so organizations should audit which document formats trigger HTML parsing and whether those documents originate from untrusted sources. Applications using Docling as a dependency should verify their current version and upgrade paths.
Exploitability
The vulnerability requires a user or automated system to process a specially crafted document, making it not immediately exploitable in passive contexts. However, if your organization processes documents from external sources—such as customer uploads, email attachments, or public repositories—the attack surface is substantial. An attacker can distribute malicious documents through common channels (email, document sharing platforms, or supply-chain vectors) and reasonably expect them to be processed. The attack does not require authentication or special privileges, only that the document be parsed by an affected Docling version.
Remediation
Upgrade Docling to version 2.94.0 or later, which addresses the unsafe URI and path handling in the HTML backend. If immediate upgrade is not possible, implement strict input validation on documents before passing them to Docling—restrict accepted file types to those not using HTML parsing, or sandbox Docling processing in isolated environments with resource limits. Monitor for unexpected resource consumption during document processing.
Patch guidance
Update your Docling dependency to 2.94.0 or newer via your package manager (pip for Python environments). Verify the update by checking your installed version and testing with a representative sample of your document workload. If Docling is vendored or pinned in your supply chain, coordinate the update with your build and deployment processes. Consult the official Docling project repository and release notes to confirm the patch is available and compatible with your Python version and other dependencies.
Detection guidance
Monitor for abnormal resource consumption (CPU, memory, disk I/O) during document processing operations. Log and review which document formats and sources trigger HTML backend processing. If possible, instrument Docling's initialization or parsing calls to capture the URI and path references it encounters, flagging any that appear malformed or reference unexpected system paths. Intrusion detection systems can flag HTTP requests attempting to deliver documents with suspicious embedded URIs, though this requires understanding your typical document sources.
Why prioritize this
This is a high-severity vulnerability with network-exploitable attack vector and user-interaction requirement. The combination of high confidentiality impact and the likelihood that many organizations process external documents elevates priority. However, organizations that only process internal, trusted documents or do not use the HTML backend can deprioritize relative to other exposures. Assess your document intake workflows and whether Docling processes untrusted sources.
Risk score, explained
The CVSS 3.1 score of 7.1 (HIGH) reflects the network-accessible attack vector, low attack complexity, and absence of privilege requirements, balanced against the user-interaction requirement and limited scope. The high confidentiality impact and moderate availability impact justify the elevated score. The score does not account for your organization's specific use case; those processing only internal documents should interpret risk as lower, while those with public document upload capabilities should treat it as more urgent.
Frequently asked questions
Do we need to upgrade if we don't process HTML documents with Docling?
The vulnerability is specific to the HTML backend's URI and path handling. If your Docling usage is limited to other formats (PDF, DOCX, etc.) and does not parse HTML, your immediate risk is lower. However, verify your actual document pipeline and whether Docling might parse HTML as an intermediate or fallback step.
Can this vulnerability be exploited without user interaction?
No. The CVSS vector indicates UI:R, meaning user interaction is required. An attacker cannot exploit this through passive network activity alone; the vulnerability is triggered when a user or application processes a malicious document. This limits mass-exploitation scenarios but does not eliminate risk if you accept external documents.
How should we handle Docling if we cannot upgrade immediately?
Implement defense-in-depth: restrict document sources to trusted partners, disable HTML parsing if not needed, process documents in isolated environments with resource caps, and monitor for unusual resource consumption. Test the upgrade thoroughly in a non-production environment before deploying widely.
Will upgrading Docling break our existing integrations?
Version 2.94.0 is a patch release addressing a security issue. Consult the Docling release notes and changelog to confirm compatibility with your Python version and dependent libraries. Test the upgrade in a development environment replicating your production setup before deployment.
This analysis is based on the published CVE record and vendor advisory information as of the modification date. CVSS scores reflect a standardized but general risk profile; your organization's actual risk depends on deployment specifics, document sources, and business context. No exploit code or attack procedures are described herein. Always verify patch availability and compatibility through official vendor sources before deploying updates. SEC.co makes no warranty regarding the completeness or accuracy of this analysis. Source: NVD (public-domain), retrieved 2026-08-05. Analysis generated by SEC.co (claude-haiku-4-5).
Related vulnerabilities
- CVE-2023-54365HIGHTraefik HTTP/2 Denial of Service Vulnerability – Rapid Reset Attack
- CVE-2024-14036HIGHDräger Core Denial of Service via Malformed SDC Messages
- CVE-2025-52293HIGHGPAC MP4Box HEVC Parser Denial of Service (CVSS 7.5)
- CVE-2025-52465HIGHGeoServer Master Password Plaintext Dump Vulnerability
- CVE-2025-53114HIGHCometD Denial-of-Service via Unacknowledged Message Queue Overflow
- CVE-2025-61025HIGHOpenLink Virtuoso 7.2.11 Remote DoS via SQL Query Parsing
- CVE-2025-71324HIGHFlowise Unauthenticated Path Traversal & Database Disclosure
- CVE-2026-10069HIGHShibby Tomato miniupnpd Resource Exhaustion Vulnerability