@inproceedings{lal-etal-2024-automated,
    title = "Automated Adversarial Discovery for Safety Classifiers",
    author = "Lal, Yash Kumar  and
      Lahoti, Preethi  and
      Sinha, Aradhana  and
      Qin, Yao  and
      Balashankar, Ananth",
    editor = "Chang, Kai-Wei  and
      Ovalle, Anaelia  and
      Zhao, Jieyu  and
      Cao, Yang Trista  and
      Mehrabi, Ninareh  and
      Galstyan, Aram  and
      Dhamala, Jwala  and
      Kumar, Anoop  and
      Gupta, Rahul",
    booktitle = "Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024)",
    month = jun,
    year = "2024",
    address = "Mexico City, Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.trustnlp-1.2",
    pages = "13--26",
    abstract = "Safety classifiers are critical in mitigating toxicity on online forums such as social media and in chatbots. Still, they continue to be vulnerable to emergent, and often innumerable, adversarial attacks.Traditional automated adversarial data generation methods, however, tend to produce attacks that are not diverse, but variations of previously observed harm types.We formalize the task of automated adversarial discovery for safety classifiers - to find new attacks along previously unseen harm dimensions that expose new weaknesses in the classifier.We measure progress on this task along two key axes (1) adversarial success: does the attack fool the classifier? and (2) dimensional diversity: does the attack represent a previously unseen harm type?Our evaluation of existing attack generation methods on the CivilComments toxicity task reveals their limitations: Word perturbation attacks fail to fool classifiers, while prompt-based LLM attacks have more adversarial success, but lack dimensional diversity.Even our best-performing prompt-based method finds new successful attacks on unseen harm dimensions of attacks only 5{\%} of the time.Automatically finding new harmful dimensions of attack is crucial and there is substantial headroom for future research on our new task.",
}