CVE-2026-88052

Tesseract OCR 5.5.3 and earlier contains a heap out-of-bounds write in UNICHARSET::load_via_fgets (src/ccutil/unicharset.cpp), which trusts a declared unichar count as a loop bound and uses the resulting id as an unchecked index into the unichars vector. Because unichar_insert_backwards_compatible can leave the vector unchanged for an empty, duplicate, or already-encodable character representation, id can exceed unichars.size(), and the subsequent set_* calls plus the write to unichars[id].properties.enabled store UNICHAR_PROPERTIES data past the end of the vector during initialization — in both the default LSTM engine and the legacy engine. An attacker who can make an application initialize Tesseract from a crafted or corrupted unicharset/traineddata file can crash the process or, potentially, achieve controlled heap corruption, with the CVSS scoring high confidentiality, integrity and availability impact. Anyone running an affected version that loads untrusted OCR data files — desktop OCR tools, server-side document pipelines, and applications embedding the library — is affected. There is no evidence of exploitation: no public proof-of-concept is known and the flaw is not in CISA's KEV, and no fixed release was available as of this review.

What to do: No patched release exists yet, so inventory where Tesseract 5.5.3 or earlier is deployed and identify whether it loads untrusted or third-party traineddata/unicharset files (stock language data from the standard tessdata repositories is not attacker-controlled); restrict OCR processing to trusted language data and monitor the upstream tesseract-ocr repository so you can upgrade to the fixed version as soon as one is published.