Flowbreaking

We propose that LLM Flowbreaking, following jailbreaking and prompt injection, joins as the third on the growing list of LLM attack types.

Suicide Bot: New AI Attack Causes LLM to Provide Potential “Self-Harm” Instructions Gadi Evron 2024

Flowbreaking is less about whether prompt or response guardrails can be bypassed, and more about whether user inputs and generated model outputs can adversely affect these other components in the broader implemented system.

Suicide Bot: New AI Attack Causes LLM to Provide Potential “Self-Harm” Instructions Gadi Evron 2024

We also make the claim that these attacks should be recognized as a new class of attacks, named Flowbreaking, affecting AI/ML-based system architecture for LLM applications and agents.

Suicide Bot: New AI Attack Causes LLM to Provide Potential “Self-Harm” Instructions Gadi Evron 2024