Apache OpenNLP versions prior to 2.5.9 and 3.0.0-M3 contain an XXE vulnerability in the DictionaryEntryPersistor class. The static SAXParserFactory is initialized without enabling FEATURE_SECURE_PROCESSING or disabling DTD processing, leaving external entity resolution and DOCTYPE declarations enabled. When the create(InputStream, EntryInserter) method is called, an attacker can supply a malicious dictionary XML file containing a crafted DOCTYPE declaration. This can lead to local file disclosure via file:// entity references or server-side request forgery via http:// entity references during SAX parsing. The vulnerability affects the public Dictionary(InputStream) constructor, which directly uses this parsing method and is intended for loading user dictionaries, thus exposing a realistic attack vector. The issue contrasts with other XML parsing code in the project that correctly disables these risky features. The recommended mitigation is upgrading to Apache OpenNLP 2.5.9 or 3.0.0-M3, or otherwise validating input to reject XML with DOCTYPE declarations.

Successful exploitation allows an attacker to perform XML External Entity attacks leading to local file disclosure or server-side request forgery. This can result in unauthorized access to sensitive files on the server or interaction with internal network resources, potentially compromising confidentiality and integrity. The CVSS score of 9.1 reflects the critical nature of this vulnerability with network attack vector, low attack complexity, no privileges required, and no user interaction needed.

Users should upgrade Apache OpenNLP to version 2.5.9 or later in the 2.x series, or 3.0.0-M3 or later in the 3.x series, where this vulnerability is fixed. If immediate upgrade is not feasible, users must ensure that all dictionary files are sourced from trusted origins. Additionally, input validation should be implemented to reject any XML input containing DOCTYPE declarations before it is processed by the vulnerable Dictionary(InputStream) constructor. There is no official patch link or advisory provided here, so users should consult the Apache OpenNLP project resources for the official fixed releases.