NLTK EUVD-2026-66397

Exploitation requires the application to pass attacker-controlled strings as the tgrep pattern argument to tgrep_positions() or tgrep_compile() - specifically, the attacker must control the pattern itself (not merely the tree being searched). The /regex/ node syntax must survive to the _tgrep_node_action function with a malicious payload intact. Deployments using NLTK tgrep exclusively with hardcoded or internally generated patterns are not exploitable regardless of network exposure. The CVSS 4.0 AT:P (attack requirement: present) and AC:H metrics formally capture this dependency on a non-default, explicitly exposed API configuration. Exploitation is limited to availability impact only - there is no code execution, data exfiltration, or integrity violation possible through this path. Authentication requirements are not confirmed from the available data; the PR:N CVSS metric implies no authentication is required once the tgrep API endpoint is reached.

The CVSS 4.0 score of 8.2 reflects High availability impact (VA:H) with zero confidentiality or integrity exposure. The vector AV:N/AC:H/AT:P confirms network reachability but flags two important mitigators: High attack complexity and a specific attack prerequisite requiring the tgrep API to be exposed to external, attacker-controlled input. EPSS data was not provided, limiting probabilistic exploitation assessment, and no CISA KEV listing exists. The absence of both KEV and public POC suggests this has not seen opportunistic or widespread exploitation. Real-world risk is highly deployment-dependent: organizations running web-facing NLP services that pass user input directly to tgrep functions face genuine denial-of-service risk, while the far more common offline batch-processing use of NLTK is not exploitable via this vector. The CVSS 4.0 score appropriately captures the High potential impact, but the AT:P requirement means security teams should audit their actual tgrep exposure before escalating priority - this is not a broad network-sweep risk.

An attacker identifies a web API endpoint that accepts tgrep search patterns and passes them without sanitization to NLTK's tgrep_positions() function. The attacker submits a crafted tgrep pattern containing a /regex/ node with a polynomial backtracking expression (e.g., a nested quantifier pattern like /(a+)+$/) as the node label matcher. Python's re.search engine processes the pattern against each tree node label, entering catastrophic backtracking that saturates a CPU core and blocks the Python worker process, rendering the service unavailable to all concurrent users.