keyword
toxicity classifier
A toxicity classifier is a natural language processing model trained to detect, flag, or score harmful, abusive, or offensive language in text. By analyzing text inputs such as online comments, conversational dialogue, or machine-generated content, the classifier estimates the likelihood or severity of toxic elements, which can include hate speech, harassment, profanity, personal attacks, and threats. Beyond serving as a tool for automated content moderation on digital platforms, toxicity classifiers are frequently employed in artificial intelligence development to evaluate safety risks in large language models, benchmark algorithmic detoxification methods, and filter unsafe content from training data.
5 items
Learning Diverse Attacks on Large Language Models for Robust Red-Teaming and Safety Tuning
Seanie Lee, Minsu Kim, Lynn Cherif, David Dobre, Juho Lee, Sung Ju Hwang, Kenji Kawaguchi, Gauthier Gidel, Yoshua Bengio, Nikolay Malkin, Moksh Jain
Why you should read this
Proposes a GFlowNet fine-tuning method to overcome mode collapse in automated red-teaming, generating diverse and transferable attack prompts that improve safety tuning for large language models against adversarial exploitation.
Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are effective against a wide range of target LLMs, both with and without safety tuning, and transfer well between target LLMs. Finally, we demonstrate that models safety-tuned using a dataset of red-teaming prompts generated by our method are robust to attacks from other RL-based red-teaming approaches.
Added
2026-10-03
License
Published with permission
ParaDetox: Detoxification with Parallel Data
Varvara Logacheva, Daryna Dementieva, Sergey Ustyantsev, Daniil Moskovskiy, David Dale, Irina Krotova, Nikita Semenov, Alexander Panchenko
Why you should read this
Presents an affordable crowdsourcing pipeline and the first parallel text detoxification datasets, demonstrating that training supervised models on paired toxic-to-neutral sentences significantly outperforms unsupervised style transfer methods.
Added
2026-10-02
ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection
Thomas Hartvigsen, Saadia Gabriel, Hamid Palangi, Maarten Sap, Dipankar Ray, Ece Kamar
Why you should read this
Introduces ToxiGen, a large-scale balanced dataset of over 274,000 machine-generated statements across 13 minority groups, along with an adversarial decoding method to help classifiers detect subtle, implicit hate speech without over-relying on identity mentions.
Toxic language detection systems often falsely flag text that contains minority group mentions as toxic, as those groups are often the targets of online hate. Such over-reliance on spurious correlations also causes systems to struggle with detecting implicitly toxic language. To help mitigate these issues, we create TOXIGEN, a new large-scale and machine-generated dataset of 274k toxic and benign statements about 13 minority groups. We develop a demonstration-based prompting framework and an adversarial classifier-in-the-loop decoding method to generate subtly toxic and benign text with a massive pretrained language model (Brown et al., 2020). Controlling machine generation in this way allows TOXIGEN to cover implicitly toxic text at a larger scale, and about more demographic groups, than previous resources of human-written text. We conduct a human evaluation on a challenging subset of TOXIGEN and find that annotators struggle to distinguish machine-generated text from human-written language. We also find that 94.5% of toxic examples are labeled as hate speech by human annotators. Using three publicly-available datasets, we show that finetuning a toxicity classifier on our data improves its performance on human-written data substantially. We also demonstrate that TOXIGEN can be used to fight machine-generated toxicity as finetuning improves the classifier significantly on our evaluation subset.
Added
2026-09-26
RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models
Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, Noah A. Smith
Why you should read this
Introduces RealToxicityPrompts, a 100,000-prompt benchmark to evaluate toxic degeneration in language models, demonstrating that benign prompts can trigger severe toxicity and that current mitigation methods remain inadequate.
Pretrained neural language models (LMs) are prone to generating racist, sexist, or otherwise toxic language which hinders their safe deployment. We investigate the extent to which pretrained LMs can be prompted to generate toxic language, and the effectiveness of controllable text generation algorithms at preventing such toxic degeneration. We create and release RealToxicityPrompts, a dataset of 100K naturally occurring, sentence-level prompts derived from a large corpus of English web text, paired with toxicity scores from a widely-used toxicity classifier. Using RealToxicityPrompts, we find that pretrained LMs can degenerate into toxic text even from seemingly innocuous prompts. We empirically assess several controllable generation methods, and find that while data- or compute-intensive methods (e.g., adaptive pretraining on non-toxic data) are more effective at steering away from toxicity than simpler solutions (e.g., banning "bad" words), no current method is failsafe against neural toxic degeneration. To pinpoint the potential cause of such persistent toxic degeneration, we analyze two web text corpora used to pretrain several LMs (including GPT-2; Radford et. al, 2019), and find a significant amount of offensive, factually unreliable, and otherwise toxic content. Our work provides a test bed for evaluating toxic generations by LMs and stresses the need for better data selection processes for pretraining.
Added
2026-09-18
Model Cards for Model Reporting
Margaret Mitchell, Simone Wu, Andrew Zaldivar, Parker Barnes, Lucy Vasserman, Ben Hutchinson, Elena Spitzer, Inioluwa Deborah Raji, Timnit Gebru
Why you should read this
Establishes a transparent documentation template that forces practitioners to explicitly declare model limitations, intended use cases, and disaggregated performance metrics across demographic subpopulations.
Trained machine learning models are increasingly used to perform high-impact tasks in areas such as law enforcement, medicine, education, and employment. In order to clarify the intended use cases of machine learning models and minimize their usage in contexts for which they are not well suited, we recommend that released models be accompanied by documentation detailing their performance characteristics. In this paper, we propose a framework that we call model cards, to encourage such transparent model reporting. Model cards are short documents accompanying trained machine learning models that provide benchmarked evaluation in a variety of conditions, such as across different cultural, demographic, or phenotypic groups (e.g., race, geographic location, sex, Fitzpatrick skin type [15]) and intersectional groups (e.g., age and race, or sex and Fitzpatrick skin type) that are relevant to the intended application domains. Model cards also disclose the context in which models are intended to be used, details of the performance evaluation procedures, and other relevant information. While we focus primarily on human-centered machine learning models in the application fields of computer vision and natural language processing, this framework can be used to document any trained machine learning model. To solidify the concept, we provide cards for two supervised models: One trained to detect smiling faces in images, and one trained to detect toxic comments in text. We propose model cards as a step towards the responsible democratization of machine learning and related artificial intelligence technology, increasing transparency into how well artificial intelligence technology works. We hope this work encourages those releasing trained machine learning models to accompany model releases with similar detailed evaluation numbers and other relevant documentation.
Added
2026-03-22