Gptline: NLTK: Entity-expansion DoS (billion laughs) via remaining raw ElementTree parses (CVE-2026-78681)

Several XML parsing sites in NLTK (<= 3.10.2) use xml.etree.ElementTree directly, which processes <!ENTITY> declarations in internal DTD subsets, allowing crafted XML documents to expand exponentially in memory (billion laughs attack). This causes a memory amplification denial-of-service (CWE-776). The vulnerability affects call sites including nltk.chunk.named_entity.load_ace_file, nltk.internals.ElementWrapper, and nltk.downloader XML parsing functions. The underlying libexpat parser added an input-amplification cap in version 2.6.0, but it only activates above ~8 MiB output and depends on the linked libexpat version; older versions have no cap. The fix routes all XML parsing through nltk.xmlsec, which forbids entity declarations and prefers defusedxml, falling back to an expat pre-scan to block malicious entities. The patch rejects malicious payloads that previously caused large memory expansions.

Exploitation of this vulnerability allows an attacker to cause a denial-of-service by consuming excessive memory through crafted XML documents with nested entity expansions. This can lead to application crashes or degraded service availability. There is no remote code execution or file disclosure as external entities are not resolved. The impact is limited to memory exhaustion and denial-of-service conditions.

A patch is available that routes all affected XML parsing through the nltk.xmlsec module, which forbids entity declarations and uses defusedxml or a fallback expat pre-scan to prevent entity expansion attacks. Users should upgrade to the fixed version >=1.0.8_23. No additional action is required if the patch is applied.