Loading vulnerability…
Docling Core (docling-core) versions >= 2.21.0 and < 2.48.4 expose a PyYAML deserialization remote code execution condition when applications invoke docling_core.types.doc.DoclingDocument.load_from_yaml() on untrusted YAML while using PyYAML versions prior to 5.4. In the vulnerable configuration, docling-core uses yaml.FullLoader, which can allow construction of unsafe Python objects during YAML parsing, enabling code execution (aligned with the PyYAML issue CVE-2020-14343). The issue is fixed in docling-core 2.48.4 by switching deserialization to yaml.SafeLoader to prevent untrusted YAML from triggering code execution.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
DoclingDocument.load_from_yaml() on attacker-controlled YAML when the environment uses PyYAML < 5.4. This can lead to full compromise of the affected service/application, including arbitrary command execution and follow-on actions consistent with the application’s privileges and accessible resources.If you can’t patch tonight, do this now.
docling_core.types.doc.DoclingDocument.load_from_yaml(). If an immediate docling-core upgrade is not possible, ensure PyYAML is upgraded to version 5.4 or greater to avoid exposure to the underlying PyYAML issue (CVE-2020-14343).Patch, then assume compromise.
yaml.FullLoader to yaml.SafeLoader, preventing unsafe object construction from untrusted YAML.1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
Repository is a PoC + helper scanner for CVE-2026-24009 (docling-core unsafe YAML deserialization) that becomes exploitable when combined with PyYAML < 5.4 behavior (referenced as CVE-2020-14343). Structure: - README.MD: explains vulnerable version ranges (docling-core >=2.21.0,<2.48.4; PyYAML <5.4), root cause (unsafe loader usage), reproduction steps, and mitigations (upgrade docling-core >=2.48.4 or PyYAML >=5.4). - malicious.yaml: payload YAML using !!python/object/new and !!python/name:eval to execute an OS command during deserialization. - repro_docling_load.py: PoC runner that calls DoclingDocument.load_from_yaml('malicious.yaml'), then checks for /tmp/docling_cve_poc_marker to confirm execution occurred during parsing (even though validation later fails). - check_loader.py: introspection utility that prints PyYAML version and the source of DoclingDocument.load_from_yaml to confirm whether yaml.FullLoader vs yaml.SafeLoader is used. - scanner/scan_cve_2026_24009.py (+ scanner/README.md): a simple local scanner that (1) queries installed docling-core and PyYAML versions by spawning a specified Python interpreter, and (2) optionally scans a project directory for direct sink usage patterns (DoclingDocument.load_from_yaml(...) or load_from_yaml(...) with import hints). It outputs classifications like NOT_VULNERABLE, VULNERABLE_DEPENDENCIES_ONLY, or POTENTIALLY_EXPLOITABLE. Overall purpose: demonstrate and validate RCE via unsafe YAML loading in docling-core under specific dependency conditions, and provide a lightweight tool to identify vulnerable environments and direct sink usage.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
6 sources tracked across advisories and community write-ups. News coverage will land here when it surfaces.
No news coverage yet. Advisories and community discussion only.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.