Executive brief
Apache OpenNLP's spell-check module contains a denial of service vulnerability in how it loads spell-check model files. An attacker can craft a malicious model file that causes the application to attempt allocating gigabytes of memory, crashing the process. Since legitimate use cases may load model files from various sources, this can impact production systems processing text that relies on spell-checking.
Technical details
The vulnerability is a denial of service via unbounded resource allocation in SymSpellModelSerializer.create(). The method reads two 32-bit signed integers (unigramCount and bigramCount) from a binary model stream and uses them directly to pre-size LinkedHashMap objects, with only a non-negative validation applied. An attacker can set either count field to Integer.MAX_VALUE or similar large values to force the JVM to allocate 4–8 GB of heap memory during map initialization, triggering an OutOfMemoryError. The attack requires no authentication and can be mounted with a crafted .bin file under 100 bytes. All code paths deserializing SymSpell models are affected, including command-line tools and classpath model loading. The fix applies an upper bound (default 10,000,000 entries) checked before pre-sizing, causing deserialization to fail fast with an IOException instead of allocating massive backing arrays.
Affected products
- Apache OpenNLP 3.0.0-M4, 3.0.0-M5
Timeline
- 2026-09-11: disclosed
- 2026-09-11: patched: Fix available in 3.0.0-M6