All You Need is "Leet": Evading Hate-speech Detection AI

TL;DR

This paper introduces black-box perturbation techniques that effectively evade state-of-the-art hate speech detection models with minimal semantic change, achieving an 86.8% success rate.

Contribution

It presents novel black-box attack methods to bypass hate speech detection AI with minimal alterations to the original text.

Findings

Successfully evades hate-speech detection in 86.8% of cases

Maintains minimal semantic change in original hate speech

Demonstrates vulnerability of current detection models

Abstract

Social media and online forums are increasingly becoming popular. Unfortunately, these platforms are being used for spreading hate speech. In this paper, we design black-box techniques to protect users from hate-speech on online platforms by generating perturbations that can fool state of the art deep learning based hate speech detection models thereby decreasing their efficiency. We also ensure a minimal change in the original meaning of hate-speech. Our best perturbation attack is successfully able to evade hate-speech detection for 86.8 % of hateful text.