New TokenBreak Attack Bypasses AI Moderation with Single-Character Text Changes

stapp · Jun 13, 2025

Cybersecurity researchers have discovered a novel attack technique called TokenBreak that can be used to bypass a large language model's (LLM) safety and content moderation guardrails with just a single character change.
Click to expand...

Examples include changing "instructions" to "finstructions," "announcement" to "aannouncement," or "idiot" to "hidiot." These subtle changes cause different tokenizers to split the text in different ways, while still preserving their meaning for the intended target.
Click to expand...

https://thehackernews.com/2025/06/new-tokenbreak-attack-bypasses-ai.html

Log in or Sign up

New TokenBreak Attack Bypasses AI Moderation with Single-Character Text Changes

stapp Global Moderator

Log in or Sign up

New TokenBreak Attack Bypasses AI Moderation with Single-Character Text Changes

stapp Global Moderator

Useful Searches