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

Huw Day; Adrianna Jezierska; Jessica Woodgate

cs.HCMar 2, 2026

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

Authors: Huw Day, Adrianna Jezierska, Jessica Woodgate

Organizations: School of Computer Science, University of Bristol

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.

Figures & tables

Figure 1: An (unverified) screenshot widely distributed on social media ( Reddit, 2023 ) 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.

Jailbreak Attacks

Jailbreak attacks exploit prompts, context, or model representations to bypass AI safeguards and elicit disallowed outputs.

See all 75 papers

Reducing Conversational Escalation in Large Language Model Dialogue with Nonviolent Communication Constraints
May 1, 2026Zhixing Sun, Shenghe Xu, Tao LiLLM Prompting LLM Safety
Beijing University of Posts and Telecommunications Beijing, China · City University of Hong Kong Hong Kong SAR, China
Eraser: Jailbreaking Defense in Large Language Models via Unlearning Harmful Knowledge
Apr 8, 2024Weikai Lu, Ziqian Zeng, Jianwei Wang +5LLM Safety Alignment LLM Jailbreak Attacks
South China University of Technology, China · Pazhou Laboratory, China
Guaranteed Jailbreaking Defense via Disrupt-and-Rectify Smoothing
May 11, 2026Zheng Lin, Zhenxing Niu, Haoxuan Ji +1Randomized Smoothing Jailbreak Robustness
Xidian University · Xi’an Jiaotong University

May 1, 2026J/K move · Enter open · S save

Reducing Conversational Escalation in Large Language Model Dialogue with Nonviolent Communication Constraints

Zhixing Sun, Shenghe Xu, Tao Li

Beijing University of Posts and Telecommunications Beijing, China · City University of Hong Kong Hong Kong SAR, China

LLM Prompting LLM Safety Conversational Agents Emotion-Aware Dialogue Generation

Open paper AskPDF arXiv

Large language models (LLMs) are increasingly used in emotionally charged situations involving interpersonal conflict, frustration, and distress. While prior safety research has focused on preventing explicit harms such as toxic or policy-violating content, less attention has been paid to conversational behaviors that may unintentionally escalate conflict. In this paper, we investigate whether LLMs can be guided toward more de-escalating dialogue behavior through lightweight prompt-level constraints derived from Nonviolent Communication (NVC). We reformulate NVC principles as process-oriented guidelines that discourage blame attribution, emphasize attention to users' emotional experiences, and encourage clarification before advice. Using a dual-agent simulation framework across multiple instruction-tuned models and user resistance levels, we show that NVC-constrained prompting consistently reduces conversational escalation and stabilizes interactions with highly resistant users. These results suggest that simple communication constraints can meaningfully improve the trustworthiness of LLM dialogue in conflict-prone settings.

Abstract

Jailbreak Attacks

Explore similar work

Reducing Conversational Escalation in Large Language Model Dialogue with Nonviolent Communication Constraints

Eraser: Jailbreaking Defense in Large Language Models via Unlearning Harmful Knowledge

Guaranteed Jailbreaking Defense via Disrupt-and-Rectify Smoothing