Original title: Detoxifying Toxic Communication: A Design Science Approach to Responsible AI
Hossein Arshadi Soufiani, Henry M. Kim, Hjalmar Turesson, Syed Mohammad Arham Noman +1 · August 2026
Rude and sarcastic messages in online workplaces hurt team trust and morale. Standard moderation tools simply delete or block these messages, which interrupts conversations without solving the core issue. To fix this, researchers built an artificial intelligence system that cleans up harmful text instead of erasing it. First, the tool scans incoming messages with specialized language models to spot toxic text. Next, a generative rewriting model transforms the offensive wording into polite, non-offensive language that keeps the original meaning intact. The team found that their tool detects toxicity with high accuracy. It also successfully preserves the sender's meaning in the rewritten messages, allowing conversations to continue smoothly while encouraging respectful communication at work.
Imagine you are refereeing a heated argument between two friends. If you just send both of them home, the argument stops but the friendship suffers and nothing gets resolved. A better referee lets each person say what they meant, but asks them to rephrase the hurtful parts. In this paper, the tool acts like that referee: it spots the hurtful words and rewrites the message so the original point comes through without the insult.
Most online moderation silences people by deleting their words. This tool offers a middle path: it rewrites toxic messages so the underlying point survives without the insult. For teams that live in chat apps, that means fewer flare-ups, less lost context, and a way to correct bad behavior without kicking anyone out.
The abstract reports technical evaluation but no details on dataset size, baseline comparisons, or real-world deployment. It does not measure long-term effects on team culture or user acceptance. The fairness claim is a design principle, not an evaluated outcome.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Detoxifying Toxic Communication: A Design Science Approach to Responsible AI — Hossein Arshadi Soufiani, Henry M. Kim, Hjalmar Turesson et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.