Tesseract can deserialize a crafted LSTM .traineddata file that contains an unsanitized signed integer multiplied during Convolve::DeSerialize. This computation incorrectly wraps the convolution output‑channel count, producing an undersized forward‑pass output buffer while subsequent writes use the unwrapped value. The result is a heap out-of-bounds write that corrupts adjacent memory, potentially causing the OCR engine to crash or leading to memory corruption.

The vulnerability affects all Tesseract‑OCR installations using versions earlier than 5.5.3, including 5.4.x and 5.5.2. Version 5.5.3 and later contain the fix and therefore are not impacted.

The CVSS score of 6.8 indicates moderate severity. EPSS data is not available, and the issue is not listed in the CISA KEV catalog. Exploitation would require an attacker to supply a malicious .traineddata file to a Tesseract instance that processes incoming OCR jobs, which could be local or remote depending on deployment. The risk is mitigated by the described patch but remains significant until updated.