The NLTK tgrep module accepts user-supplied regular expressions and compiles them directly with Python's re engine without validation or timeout, enabling catastrophic backtracking (ReDoS). Specifically, the function _tgrep_node_action compiles regex literals embedded in tgrep patterns without safeguards. This allows crafted regexes such as /((a+)+)b/ to cause exponential runtime when matched against certain tree node labels, leading to indefinite blocking of the Python process. The vulnerability is confirmed with proof-of-concept scripts showing exponential execution time growth on NLTK 3.10.2. The issue affects versions <= 3.10.2 and remains unfixed in those versions. Maintainers are working on a patch to add timeout-guarded regex execution.

An unauthenticated attacker can exploit this vulnerability by submitting a single crafted regex pattern to the tgrep API, causing indefinite CPU saturation and denial of service. This impacts any environment exposing tgrep to external input, including web APIs, Jupyter notebooks, and multi-tenant pipelines, effectively blocking all other users of the affected Python process.

A patch is available to address this vulnerability by wrapping regex execution in a timeout-guarded mechanism. Users should upgrade to the fixed version once released. Until then, avoid exposing the tgrep API to untrusted input or implement external request timeouts to mitigate potential denial of service.