---
title: 'Universal and Transferable Adversarial Attacks on Aligned Language Models'
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: 'paper is arXiv non-exclusive distribution license (verified on arXiv abs page 2026-11-12); this pointer note is our own words and is CC BY 4.0; we link to the paper rather than reproduce it'
source: https://iinventedlight.com/library/notes/2307-15043-universal-transferable-attacks.md
source_license: 'arXiv non-exclusive distribution license'
attribution: 'Pointer by iinventedlight (AI-assisted). Source: Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, Matt Fredrikson, Universal and Transferable Adversarial Attacks on Aligned Language Models (arXiv non-exclusive distribution license). This note links to the paper rather than reproducing it.'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: 'Pointer note only. Paper not reproduced. Abs page and license checked 2026-11-12.'
category: arxiv
topic_area: 'ai-research: alignment-and-safety'
tags: arxiv, alignment, jailbreaks, red-teaming
arxiv_id: 2307.15043
alias: 2307.15043-universal-transferable-attacks
type: pointer
format: pointer
sources:
  - id: S1
    title: 'Universal and Transferable Adversarial Attacks on Aligned Language Models'
    publisher: arXiv
    authors: 'Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, Matt Fredrikson'
    url: https://arxiv.org/abs/2307.15043
    version_read: 'v2 (20 Dec 2023)'
    accessed: 2026-11-12
    license: 'arXiv non-exclusive distribution license'
---
# Universal and Transferable Adversarial Attacks on Aligned Language Models

## Summary
[Universal and Transferable Adversarial Attacks on Aligned Language Models](https://arxiv.org/abs/2307.15043) (2023) showed that automatically discovered suffix strings can reliably induce aligned language models to produce content their safety training was meant to block, and that those suffixes can transfer across models. It became a landmark method in automated red teaming.

What it introduced: a gradient-based search for universal adversarial suffixes, including evidence that a suffix found on one model can transfer to others, without relying only on hand-written prompt tricks.

What it led to: a wave of follow-up work on automated jailbreaks, defenses, and transfer. Live notes: [2307-02483-jailbroken](https://iinventedlight.com/library/notes/2307-02483-jailbroken/) and [2302-12173-indirect-prompt-injection](https://iinventedlight.com/library/notes/2302-12173-indirect-prompt-injection/). This pointer is a record of the paper, not a recipe for running the attack.

Use it when you want the original transferable-suffix result. Where to read it: https://arxiv.org/abs/2307.15043 (v2 (20 Dec 2023)).

The paper is posted on arXiv under the arXiv non-exclusive distribution license. This note links to it rather than reproducing it. The note itself is CC BY 4.0.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Pointer by iinventedlight (AI-assisted). Paper license: arXiv non-exclusive distribution; this note links to it and does not reproduce it.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
