What's in a Name? Morphological Shortcuts by LLMs in Pharmacology

2026-06-03

AI Summary

This study addresses a critical safety concern in pharmacological reasoning by large language models (LLMs): their overreliance on morphological suffix cues in drug names, which leads to confidently incorrect yet plausible clinical judgments about fictitious compounds. To investigate this, the authors construct realistic fictional drug names based on authentic pharmacological suffixes and introduce the first framework for quantitatively analyzing model dependence on suffixes, stems, and holistic name semantics. Integrating behavioral experiments, semantic decomposition, and cross-model activation patching, the work systematically reveals LLMs’ implicit reliance on suffix-based heuristics. Evaluated across 653 drugs, the findings demonstrate that models predominantly infer pharmacological class from suffixes—a behavior localized to early-to-mid network layers—and rarely acknowledge this dependency, often conflating properties of drugs sharing common suffixes.

Abstract

The morphological form of a word can often give cues to its meaning, but purely relying on these mappings can lead to overgeneralization in high-stakes domains. In the medical domain, for instance, LLMs can confidently reason about fictitious drugs from their affixes alone (e.g., wugcillin) and generate plausible-looking clinical content. We present a behavioral and mechanistic study of LLM "affix heuristics" in pharmacology. Using fictitious drug names built from real affixes, we show that affix signals alone elicit class-level pharmacological responses. We introduce a framework for identifying whether a model's drug semantics are driven mainly by the affix, the stem, or the drug name as a whole. Applied across 653 drugs, our framework reveals that models often induce drug meaning primarily through affix cues, yet rarely explicitly indicate this reliance, and sometimes incorrectly conflate properties among affix-sharing drugs. Activation patching across models further localizes this behavior to early-mid layers. These findings show that morphological shortcuts pose a subtle but measurable risk to safety.

Problem

• morphological shortcuts
• affix heuristics
• large language models
• pharmacology
• drug naming

Innovation

• morphological shortcuts
• affix heuristics
• activation patching
• fictitious drug names
• pharmacological semantics