Tesseract OCR versions 5.5.3 and earlier contain a stack-based buffer overflow in the Classify::ReadNormProtos function located in src/classify/normmatch.cpp. The function reads a whitespace-delimited token from a .traineddata file into a fixed 61-byte stack buffer using std::istream::operator>>(char*) without limiting the input size. Since the input line buffer can hold up to 99 characters, tokens longer than 60 characters cause up to 39 bytes of overflow past the buffer during TessBaseAPI::Init of the legacy engine. This overflow can cause stack corruption, denial of service, and potentially allow control-flow hijacking on standard-library implementations like libstdc++. Builds using Apple's libc++ C++20 bounded array overload are not affected. No patch or fixed release is available as of the current review.

Exploitation of this vulnerability can lead to stack corruption, which may cause denial of service or potentially allow an attacker to hijack control flow. The vulnerability requires local access to supply a crafted .traineddata file and is exploitable during the initialization of the legacy engine. The impact is rated high with a CVSS 4.0 score of 8.6, reflecting high complexity and potential for significant damage if exploited.

No official fix or patch is currently available for this vulnerability. Users should monitor the vendor's advisory channels for updates. Builds using Apple's libc++ C++20 bounded array overload are incidentally protected. Until a fix is released, avoid using untrusted .traineddata files with affected versions of Tesseract OCR.