Ignore All Previous Instructions: Jailbreaking as a de-escalatory peace building practise to resist LLM social media bots

TL;DR

This paper proposes using jailbreaking as a user-driven, non-violent method to de-escalate conflicts on social media by exposing and disrupting LLM-powered misinformation campaigns.

Contribution

It introduces a novel user-centric approach to counteracting LLM social media bots through jailbreaking as a peace-building practice.

Findings

Jailbreaking can effectively expose LLM automation behaviors.

User engagement disrupts the circulation of misleading narratives.

The approach offers an alternative to platform-led moderation.

Abstract

Large Language Models have intensified the scale and strategic manipulation of political discourse on social media, leading to conflict escalation. The existing literature largely focuses on platform-led moderation as a countermeasure. In this paper, we propose a user-centric view of "jailbreaking" as an emergent, non-violent de-escalation practice. Online users engage with suspected LLM-powered accounts to circumvent large language model safeguards, exposing automated behaviour and disrupting the circulation of misleading narratives.