Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Huw Day, Adrianna Jezierska, Jessica Woodgate

TL;DR

A user-centric view of jailbreaking is proposed as an emergent, non-violent de-escalation practice in large language model safeguards, exposing automated behaviour and disrupting the circulation of misleading narratives.

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.

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

TL;DR

A user-centric view of jailbreaking is proposed as an emergent, non-violent de-escalation practice in large language model safeguards, exposing automated behaviour and disrupting the circulation of misleading narratives.

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.
Paper Structure (12 sections, 1 figure)

This paper contains 12 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Social Media, LLMs, and Political Discourse
  3. Large Language Models and their Role on Social Media.
  4. LLMs as State Funded Force Multipliers for Misinformation Operations.
  5. Escalation Through Misattributed Intent.
  6. Platform Level Countermeasures.
  7. Jailbreaking as an Emergent User Practice
  8. What is Jailbreaking?
  9. Jailbreaking in the Wild: An Example Scenario.
  10. Jailbreaking as Emergent Peace Building.
  11. Conclusion
  12. Limitations and Future Work.

Figures (1)

  • Figure 1: An (unverified) screenshot widely distributed on social media PromptInjectionExample where a user with a Russian flag as their profile picture distributes misinformation on the Russia-Ukraine war but is revealed to be an LLM by a prompt injection attack.