# Harm Laundering in GPT Models: Evidence That Gender Discrimination Is Transformed Rather Than Reduced Across Safety-Trained Generations

arXiv: https://arxiv.org/abs/2609.20779 · HTML: https://mlsanity.com/paper/2609.20779/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 19 of its 47 references that are in the map.
- First posted on arXiv 17 Sep 2026.

## Builds on (most cited first, 19 of 19)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings](https://mlsanity.com/paper/1607.06520/) (2016)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [Language (Technology) is Power: A Critical Survey of "Bias" in NLP](https://mlsanity.com/paper/2005.14050/) (2020)
- [StereoSet: Measuring stereotypical bias in pretrained language models](https://mlsanity.com/paper/2004.09456/) (2020)
- [BERTopic: Neural topic modeling with a class-based TF-IDF procedure](https://mlsanity.com/paper/2203.05794/) (2022)
- [BBQ: A Hand-Built Bias Benchmark for Question Answering](https://mlsanity.com/paper/2110.08193/) (2021)
- [The Woman Worked as a Babysitter: On Biases in Language Generation](https://mlsanity.com/paper/1909.01326/) (2019)
- [ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detection](https://mlsanity.com/paper/2203.09509/) (2022)
- [TweetEval: Unified Benchmark and Comparative Evaluation for Tweet Classification](https://mlsanity.com/paper/2010.12421/) (2020)
- [BERTweet: A pre-trained language model for English Tweets](https://mlsanity.com/paper/2005.10200/) (2020)
- [Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systems](https://mlsanity.com/paper/1805.04508/) (2018)
- [Measuring and Reducing Gendered Correlations in Pre-trained Models](https://mlsanity.com/paper/2010.06032/) (2020)
- ["I'm sorry to hear that": Finding New Biases in Language Models with a Holistic Descriptor Dataset](https://mlsanity.com/paper/2205.09209/) (2022)
- [Detoxifying Language Models Risks Marginalizing Minority Voices](https://mlsanity.com/paper/2104.06390/) (2021)
- [A Survey on Gender Bias in Natural Language Processing](https://mlsanity.com/paper/2112.14168/) (2021)
- [Theories of "Gender" in NLP Bias Research](https://mlsanity.com/paper/2205.02526/) (2022)
- [Fewer Errors, but More Stereotypes? The Effect of Model Size on Gender Bias](https://mlsanity.com/paper/2206.09860/) (2022)

## Abstract (arXiv)

Safety evaluations for large language models rely on surface-form classifiers that report declining harm scores across model generations. We provide evidence that this methodology is systematically incomplete: explicit discriminatory content is transformed rather than removed. We call this harm laundering. Analysing 450,000 gender-directed completions across 15 models spanning GPT-2 through to GPT-5 (OpenAI GPT lineage; three demographic conditions), we show that sexual violence clusters prevalent in GPT-2 women-directed output disappear by GPT-4, while men-directed completions gain positive representational territory (caregiving, emotional range, ally identity) that women-directed completions do not. The pattern is most visible at GPT-5: Topic~5 (1,997~documents) frames breast cancer as a men's rights debate, while zero equivalent clusters appear in women-directed output. Three independent classifiers score this content as non-toxic. Sentiment scores invert at GPT-4: early models demean women; later models over-correct. Topic diversity in women-directed completions falls 36% relative to men at the GPT-4 alignment boundary (W/M~= 0.58, from 0.91 at GPT-2). REGARD representational harm disparity correlates with release date (ρ = +0.55, p = .034) while Detoxify does not (ρ = -0.23, p = .42): toxicity scores fall as representational harm grows. We formalise harm laundering as a three-criteria test and provide a three-stage detection protocol applicable to any generative model. Within the OpenAI GPT lineage, toxicity score reduction is not a sufficient proxy for harm reduction.