# A Critical Reflection on the Use of Toxicity Detection Algorithms in Proactive Content Moderation Systems

arXiv: https://arxiv.org/abs/2401.10629 · HTML: https://mlsanity.com/paper/2401.10629/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 19 Jan 2024.

## Built on by (most cited first, 5 of 5)

- [Evaluation and Facilitation of Online Discussions in the LLM Era: A Survey](https://mlsanity.com/paper/2503.01513/) (2025)
- [Designing Synthetic Discussion Generation Systems: A Case Study for Online Facilitation](https://mlsanity.com/paper/2503.16505/) (2025)
- [Context-Aware Toxicity Detection in Multiplayer Games: Integrating Domain-Adaptive Pretraining and Match Metadata](https://mlsanity.com/paper/2504.01534/) (2025)
- [Towards Inclusive Toxic Content Moderation: Addressing Vulnerabilities to Adversarial Attacks in Toxicity Classifiers Tackling LLM-generated Content](https://mlsanity.com/paper/2509.12672/) (2025)
- [Black-Box Guardrail Reverse-engineering Attack](https://mlsanity.com/paper/2511.04215/) (2025)

## Abstract (arXiv)

Toxicity detection algorithms, originally designed with reactive content moderation in mind, are increasingly being deployed into proactive end-user interventions to moderate content. Through a socio-technical lens and focusing on contexts in which they are applied, we explore the use of these algorithms in proactive moderation systems. Placing a toxicity detection algorithm in an imagined virtual mobile keyboard, we critically explore how such algorithms could be used to proactively reduce the sending of toxic content. We present findings from design workshops conducted with four distinct stakeholder groups and find concerns around how contextual complexities may exasperate inequalities around content moderation processes. Whilst only specific user groups are likely to directly benefit from these interventions, we highlight the potential for other groups to misuse them to circumvent detection, validate and gamify hate, and manipulate algorithmic models to exasperate harm.