Research
[PDF] Spotting Slopsquatting Risks Caused by Package Name Hallucinations in ...
Publication date not yet evaluated · Added · 2 reports arxiv.org
Page published
Coverage timeline
Why it matters
Slopsquatting turns LLM package-name hallucinations into installable supply-chain compromises, and this work measures how often local coding models fabricate names an attacker could preregister on PyPI while offering a locally-deployable detection approach.
A research paper, 'Names Can Hurt', studies slopsquatting risks from Python package-name hallucinations in small, locally-run open-weight coding LLMs (7B-70B) and proposes a two-layer detector combining a deterministic PyPI existence check with a per-LLM Random Forest classifier, embedded in a LangGraph retry/fallback pipeline. Across 300 prompts it produces hallucination-free code on 76% of runs and finds hallucination rates scale with prompt adversariality (40-73% on slopsquat baits), with released code, prompts, and classifiers.