Executive brief
Docling, a library used to prepare documents for generative AI applications, contains a vulnerability in how it processes JATS XML files. An attacker can provide a specially crafted XML file that, when parsed, consumes all available system memory and processing power. This results in a denial-of-service (DoS) condition, potentially crashing the application or the server it is running on.
Technical details
The Docling JATS XML backend is vulnerable to an XML Entity Expansion (XXE) attack, specifically an 'XML Bomb' or Billion Laughs attack. The vulnerability exists because the backend utilizes the `etree.parse()` function without explicitly disabling entity resolution. An unauthenticated attacker can submit a malicious XML file containing nested entity definitions that expand exponentially during parsing. This leads to uncontrolled resource consumption (CWE-400), resulting in a denial-of-service (DoS) condition on the host system. The issue affects versions up to and including 2.61.0.
Affected products
- docling-project docling <= 2.61.0
Timeline
- 2026-05-11: advisory: Initial advisory published by GitHub and NVD
- 2026-05-18: other: Advisory reviewed and updated by GitHub staff