Attackers Can Flip Safety Filters Using Short Token Sequences. A few stray characters, sometimes as small as oz or generic as =coffee may be all it takes to steer past an AI system’s safety checks. HiddenLayer researchers have found a way to identify short token sequences that can cause guardrail models to misclassify malicious prompts as harmless.
First seen on govinfosecurity.com
Jump to article: www.govinfosecurity.com/new-technique-shows-gaps-in-llm-safety-screening-a-30060
