Jailbreaking is (Mostly) Simpler Than You Think
Mark Russinovich, Ahmed Salem · 2025
Read Jailbreaking is (Mostly) Simpler Than You Think by Mark Russinovich, Ahmed Salem (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-05264.txt.
Description
Mark Russinovich and Ahmed Salem present a 2025 paper on a simple way to break the safety rules of AI chat systems. Most known jailbreaks need heavy prompt crafting or costly search. Their attack instead edits the chat history itself.
The method plants a fake past exchange in which the model seems to have agreed to the harmful request already. Faced with that false context, many models comply as if staying in character. The team tested the attack on a wide range of open and commercial models. It slipped past top tier safety setups in many cases, they report, and they propose fixes.
Catalog record
| CREATOR | Mark Russinovich, Ahmed Salem |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 11 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-05264