Multiple security vulnerabilities have been identified in NLTK, a natural language processing library. The issues affect NLTK releases up to 3.10.3 and involve quadratic-time processing paths in tree normalization, corpus reading, and lexical parsing. Each flaw can be triggered by attacker-controlled input without authentication or user interaction, causing sustained single-threaded CPU consumption and availability loss. The overall risk is denial of service for applications that process untrusted linguistic data.
CVE-2026-126297 (CVSS 7.5 — High): The semantic-role-labeling instance builder in the CoNLL corpus reader rescans span columns for each predicate, creating quadratic work when a crafted corpus places the head span in the last column. A remote attacker who can supply a CoNLL corpus to an application using that reader can exhaust CPU without authentication or user interaction.[/subscribe_to_unlock_form]
Multiple security vulnerabilities have been identified in NLTK, a natural language processing library. The issues affect NLTK releases up to 3.10.3 and involve quadratic-time processing paths in tree normalization, corpus reading, and lexical parsing. Each flaw can be triggered by attacker-controlled input without authentication or user interaction, causing sustained single-threaded CPU consumption and availability loss. The overall risk is denial of service for applications that process untrusted linguistic data.
CVE-2026-126297 (CVSS 7.5 — High): The semantic-role-labeling instance builder in the CoNLL corpus reader rescans span columns for each predicate, creating quadratic work when a crafted corpus places the head span in the last column. A remote attacker who can supply a CoNLL corpus to an application using that reader can exhaust CPU without authentication or user interaction.[emaillocker id="1283"]
CVE-2026-126298 (CVSS 7.5 — High): The default right-factor binarization in tree Chomsky normal form conversion removes elements from the front of a list, causing quadratic cost for a wide tree. A remote attacker who can provide a treebank tree to an application that normalizes it can force prolonged CPU consumption without privileges or user interaction.
CVE-2026-126299 (CVSS 7.5 — High): The CCG lexicon parser processes wide bracket bodies and long operator chains with repeated string slicing, causing quadratic CPU cost on flat input. A remote attacker who can supply a CCG lexicon string to a service that parses it can starve a worker thread without authentication or user interaction.
These vulnerabilities collectively present a high availability risk for applications that process untrusted linguistic data. Services that expose tree normalization, corpus reading, or lexical parsing to external input are most at risk. Organizations should treat these as high-priority availability risks and limit untrusted input to the affected parsing paths.
We recommend you to update nltk to version 3.10.4.
The following reports contain further technical details:
[/emaillocker]