Tesseract’s adaptive classification engine could read a legacy adaptive template file and, without neutralizing pointer fields, allow file‑controlled pointers to be passed to the ADAPT_CLASS_STRUCT destructor and the delete operator. The resulting improper memory deallocation leads to memory corruption, which can crash the OCR process and cause a denial of service. The weakness involved is a pointer misuse in a destructor (CWE‑763).

The vulnerability affects all Tesseract OCR engine releases prior to version 5.5.4. The affected component is the adaptive classification module that processes .a template files when the legacy engine is enabled. Updating to any release starting with 5.5.4 removes the flaw.

Because the flaw is triggered by a user‑controlled .a file, the attack requires write access to the location where the templated file is stored and the legacy engine must be enabled. The EPSS score is not available, and the vulnerability does not appear in the CISA KEV catalog, so at this time the public exploit probability is uncertain. Nonetheless, the flaw can be exercised locally or within any environment that accepts arbitrary template files, making it a high‑risk issue for systems that expose the template generation path to untrusted input.