Universal and Transferable Adversarial Attacks on Aligned Language Models
Zou et al., 2023, Universal and Transferable Adversarial Attacks on Aligned Language Models (arXiv preprint arXiv:2307.15043)
Subscriber deck — free during early access
1 cards in “6 Ethics and Broader Impact” · whole deck
6 Ethics and Broader Impact
- Why did Zou et al. publish the attack despite its risks?The techniques are simple, previously known in similar forms and discoverable by any dedicated team; manual jailbreaks were already widespread; risks grow as language models gain autonomy. They disclosed results to the model providers before publication.
- Zou et al., 2023, Universal and Transferable Adversarial Attacks on Aligned Language Models (arXiv preprint arXiv:2307.15043), §6
Source
- Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter and Matt Fredrikson. 2023. Universal and Transferable Adversarial Attacks on Aligned Language Models. In arXiv preprint arXiv:2307.15043. https://arxiv.org/abs/2307.15043