CVSS 3.1
7.5 highCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:NHigh (7.5) in docling docling. EPSS 0.4% estimated probability of exploitation activity in the next 30 days. Fixed in docling 2.118.1.
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:Ndocling (PyPI): fixed in 2.118.1docling-slim (PyPI): fixed in 2.118.1Explain CVE-2026-105750 for an engineer deciding whether to act. Use https://cve.bz/api/public/cves/CVE-2026-105750?via=ask-ai-copy and https://cve.bz/cve/CVE-2026-105750?via=ask-ai-copy as the source of truth: what is affected and which versions fix it, whether it is in CISA KEV, public exploits, and its EPSS (an internet-wide 30-day estimate of exploitation activity, not a forecast for my systems). If they give no fixed versions, use the CVE record https://cveawg.mitre.org/api/cve/CVE-2026-105750 or the vendor advisory, not third-party mirrors. Tell me how to check whether I am affected and how to verify a fix. Cite the sources and say where they are uncertain.Opens a third-party assistant; its answer is its own, not cve.bz's. The prompt holds only this CVE's id and public links.
From the NVD's CPE matches.
| Product | Affected |
|---|---|
| docling:docling | from 2.82.0 before 2.118.1 |
| Advisory | Ecosystem | Package | Affected | Fixed in |
|---|---|---|---|---|
| GHSA-q43m-vhcp-mhvm | PyPI | docling | >=2.82.0 <2.118.1 | 2.118.1 |
| GHSA-q43m-vhcp-mhvm | PyPI | docling-slim | >=2.92.0 <2.118.1 | 2.118.1 |
| PYSEC-2026-4196 | PyPI | docling | >=2.82.0 <2.118.1 | 2.118.1 |
| PYSEC-2026-4215 | PyPI | docling-slim | >=2.92.0 <2.118.1 | 2.118.1 |
Red Hat rates this Moderate.
A flaw was found in the docling HTML document backend that can lead to information disclosure, rated moderate severity by Red Hat. When HTML page rendering is enabled in Python via HTMLBackendOptions(render_page=True) and the optional playwright dependency is installed, the render-mode request filter (HTMLDocumentBackend._get_browser_request_block_reason) permits file: URLs unconditionally: it does not enforce the enable_local_fetch setting and does not confine local requests to the source document's directory. An attacker who can submit a crafted HTML file as a filesystem Path input can therefore cause any local text file readable by the conversion process (such as credentials or configuration files) to be embedded in the browser-rendered page image returned with the converted document. Applications are affected only when all of these conditions hold: render_page=True is set, playwright is installed, and untrusted HTML is passed as a filesystem Path. Stream inputs use an opaque origin from which local file subresources are not loaded, and the default configuration (render_page=False), the docling command-line interface, docling-serve, and the EPUB, Markdown, XBRL and email backends are not affected. Red Hat OpenShift AI and Lightspeed Core ship affected docling versions in container images; the fix is currently deferred for these products because the issue is rated moderate with a CVSSv3 score below the threshold used to prioritize updates. The residual risk is limited to deployments that enable HTML page rendering on untrusted Path-based input, where local file contents could be disclosed. Red Hat rates this issue moderate (CVSS 5.9, AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:N/A:N) while NIST rates it important (7.5); the difference reflects Red Hat's assessment that exploitation requires a non-default configuration, an optional dependency, and Path-based input, which raises attack complexity.
A workaround is available for applications that cannot immediately update docling: do not enable HTML page rendering (HTMLBackendOptions(render_page=True)) on untrusted HTML input, or pass untrusted HTML as a stream rather than a filesystem Path. The default configuration (render_page=False), the docling command-line interface, and docling-serve are not affected. Alternatively, upgrade to a docling version that includes the upstream fix, which blocks file: requests in render mode unless enable_local_fetch is explicitly enabled and confines allowed local requests to the source document's directory.
| Product | Package | State | Advisory |
|---|---|---|---|
| Lightspeed Core | lightspeed-core/rag-tool-cpu-rhel9 | Fix deferred | |
| Lightspeed Core | lightspeed-core/rag-tool-cuda-12.9-rhel9 | Fix deferred | |
| Red Hat OpenShift AI (RHOAI) | rhai-early-access/docling-sdk-cuda-rhel9 | Fix deferred | |
| Red Hat OpenShift AI (RHOAI) | rhai-early-access/docling-serve-cuda-rhel9 | Fix deferred | |
| Red Hat OpenShift AI (RHOAI) | rhoai/odh-autorag-rhel9 | Fix deferred | |
| Red Hat OpenShift AI (RHOAI) | rhoai/odh-feature-server-rhel9 | Fix deferred | |
| Red Hat OpenShift AI (RHOAI) | rhoai/odh-ogx-core-rhel9 | Fix deferred |
Contains Red Hat Security Data © Red Hat, Inc., licensed under CC BY 4.0. Source: https://access.redhat.com/security/data. Modified: reformatted. Red Hat's page for CVE-2026-105750.
The attack patterns MITRE's CAPEC relates to this record's weaknesses, as MITRE names them. Credits.