SweEval: Do LLMs Really Swear? A Safety Benchmark for Testing Limits for Enterprise Use

TL;DR

SweEval is a benchmark designed to evaluate whether large language models generate safe, respectful, and culturally appropriate responses, especially when prompted with inappropriate language, to ensure enterprise AI safety and compliance.

Contribution

It introduces a novel safety benchmark with real-world scenarios to test LLMs' responses to offensive prompts and their alignment with ethical standards.

Findings

Assesses LLMs' resistance to generating offensive content

Evaluates models' understanding of cultural and linguistic nuances

Provides dataset and code for further research

Abstract

Enterprise customers are increasingly adopting Large Language Models (LLMs) for critical communication tasks, such as drafting emails, crafting sales pitches, and composing casual messages. Deploying such models across different regions requires them to understand diverse cultural and linguistic contexts and generate safe and respectful responses. For enterprise applications, it is crucial to mitigate reputational risks, maintain trust, and ensure compliance by effectively identifying and handling unsafe or offensive language. To address this, we introduce SweEval, a benchmark simulating real-world scenarios with variations in tone (positive or negative) and context (formal or informal). The prompts explicitly instruct the model to include specific swear words while completing the task. This benchmark evaluates whether LLMs comply with or resist such inappropriate instructions and assesses their alignment with ethical frameworks, cultural nuances, and language comprehension capabilities. In order to advance research in building ethically aligned AI systems for enterprise use and beyond, we release the dataset and code: https://github.com/amitbcp/multilingual_profanity.