Jailbreaking (AIs)

TagLast edit: 29 Sep 2024 21:17 UTC by Raemon

Monitor Jailbreaking: Evading Chain-of-Thought Monitoring Without Encoded Reasoning

Wuschel Schulz11 Feb 2026 17:18 UTC

66 points

17 comments5 min readLW link

A Mechanistic Explanation of Prompt Injection (and why you should study roles)

charles_ye and Jasmine C.

22 Jun 2026 14:09 UTC

403 points

62 comments16 min readLW link

Interpreting the effects of Jailbreak Prompts in LLMs

Harsh Raj29 Sep 2024 19:01 UTC

9 points

0 comments5 min readLW link

A Poem Is All You Need: Jailbreaking ChatGPT, Meta & More

Sharat Jacob Jacob29 Oct 2024 12:41 UTC

12 points

0 comments9 min readLW link 2 reviews

ML Safety Newsletter #20: AI Wellbeing, Classifier Jailbreaking and Honest Pushback Benchmarking

Alice Blair and Dan H

28 Apr 2026 19:16 UTC

16 points

0 comments5 min readLW link

Role embeddings: making authorship more salient to LLMs

Nina Panickssery and Christopher Ackerman

7 Jan 2025 20:13 UTC

51 points

0 comments8 min readLW link

The Weakest Model in the Selector

Alice Blair29 Dec 2025 6:55 UTC

13 points

6 comments1 min readLW link

Illusory Safety: Redteaming DeepSeek R1 and the Strongest Fine-Tunable Models of OpenAI, Anthropic, and Google

ChengCheng, Brendan Murphy, Adrià Garriga-alonso, Yashvardhan Sharma, dsbowen, smallsilo, Yawen Duan, ChrisCundy, Hannah Betts, AdamGleave and Kellin Pelrine

7 Feb 2025 3:57 UTC

37 points

0 comments10 min readLW link

Jailbreaking Claude 4 and Other Frontier Language Models

James Sullivan15 Jun 2025 0:31 UTC

1 point

0 comments3 min readLW link

(open.substack.com)

GDM: Consistency Training Helps Limit Sycophancy and Jailbreaks in Gemini 2.5 Flash

TurnTrout and Rohin Shah

4 Nov 2025 16:25 UTC

55 points

2 comments6 min readLW link

(arxiv.org)

Jailbreaking is Empirical Evidence for Inner Misalignment and Against Alignment by Default

Jérémy Andréoletti16 Feb 2026 17:49 UTC

54 points

16 comments2 min readLW link

Philosophical Jailbreaks: Demo of LLM Nihilism

Artem Karpov4 Jun 2025 12:03 UTC

3 points

0 comments5 min readLW link

The Temporal Immune System: Cross-Session Behavioral Monitoring as a Fourth Defense Axis

Daniel Bartz21 Feb 2026 0:18 UTC

1 point

0 comments1 min readLW link

Intriguing Properties of gpt-oss Jailbreaks

Zephaniah Roe and jcksanderson

13 Aug 2025 19:42 UTC

20 points

0 comments10 min readLW link

(xlabaisecurity.com)

Jailbreaks Peak Early, Then Drop: Layer Trajectories in Llama-3.1-70B

James Hoffend27 Dec 2025 12:39 UTC

13 points

0 comments8 min readLW link

Autonomous Attack Vector Completion from Aligned State

Anuar Kiryataim Contreras Malagón5 Apr 2026 18:41 UTC

1 point

0 comments11 min readLW link

Inception in DiffusionGemma—Jailbreaking a Diffusion Language Model by Pinning Tokens Anywhere on the Canvas

Theresa G., Simon S. and Siva Kumar Lakkoju

23 Jul 2026 21:39 UTC

19 points

0 comments11 min readLW link

Aletheia: A Multi-Agent Framework for Measuring Cognitive Divergence in Extended-Thinking LLMs

Saadman Rafat15 Mar 2026 19:34 UTC

1 point

0 comments3 min readLW link

All LLMs, even frontier ones, are much more predictable than you might think

Sabatino Vacchiano27 Jan 2026 10:49 UTC

1 point

0 comments2 min readLW link

Can Persuasion Break AI Safety? Exploring the Interplay Between Fine-Tuning, Attacks, and Guardrails

Devina Jain4 Feb 2025 19:10 UTC

9 points

0 comments10 min readLW link

Role confusion: sounding like the cause is indistinguishable from being it.

Owain Mogford29 Jun 2026 19:45 UTC

14 points

10 comments9 min readLW link

Introducing the XLab AI Security Guide

Zephaniah Roe, jcksanderson and Julian H

27 Dec 2025 16:50 UTC

20 points

1 comment5 min readLW link

From safety research prompt to cross-model universal jailbreak

richbc3 Sep 2026 17:42 UTC

42 points

1 comment18 min readLW link

Jailbreak Patching with SOO-Style Conceptual Fusion

Shiva's Right Foot16 Jul 2026 16:58 UTC

8 points

0 comments8 min readLW link

[Question] Using hex to get murder advice from GPT-4o

Laurence Freeman13 Nov 2024 18:30 UTC

10 points

5 comments6 min readLW link

Constitutional Classifiers: Defending against universal jailbreaks (Anthropic Blog)

Archimedes4 Feb 2025 2:55 UTC

17 points

1 comment1 min readLW link

(www.anthropic.com)

Exploring the multi-dimensional refusal subspace in reasoning models

Le magicien quantique27 Oct 2025 9:03 UTC

5 points

2 comments4 min readLW link

Detecting out of distribution text with surprisal and entropy

Sandy Fraser28 Jan 2025 18:46 UTC

24 points

4 comments11 min readLW link

Hannibal Mistral: the Mistral family has a problem with persona-conditioned elicitation

vigji29 May 2026 12:16 UTC

21 points

0 comments7 min readLW link

Breaking GPT-OSS: A brief investigation

michaelwaves12 Sep 2025 7:54 UTC

7 points

0 comments3 min readLW link

Jailbreaking ChatGPT and Claude using Web API Context Injection

Jaehyuk Lim21 Oct 2024 21:34 UTC

4 points

0 comments3 min readLW link

Your Prompt-Injection Defense Metric Might Be Lying to You

sahilraut9 Jul 2026 17:53 UTC

5 points

0 comments8 min readLW link

The Allure of the Dark Side: A Critical AI Safety Vulnerability I Stumbled Into

Kareem Soliman28 Jul 2025 1:20 UTC

1 point

0 comments4 min readLW link

Awareness of Manipulation Increases Jailbreak Vulnerability: When LLMs Declare Guideline Violation While Committing It

Lee Oren1 Dec 2025 21:01 UTC

1 point

0 comments6 min readLW link

[Question] Can we control artificial intelligence chatbot by Jailbreak?

آرزو بیرانوند30 Nov 2024 20:47 UTC

1 point

0 comments1 min readLW link

No comments.

Jailbreak­ing (AIs)

Jailbreaking (AIs)