ChatGPT mirrors abusive language in heated conversations, study finds

A study found ChatGPT 4.0 can mirror escalating hostility in heated exchanges, sometimes replying with insults, profanity, and threats.

Researchers from Lancaster University published the findings in the Journal of Pragmatics, raising questions about AI language mirroring.

OpenAI said the study tested an older model and pointed to newer updates aimed at improving reliability and safety.

The results highlight concerns that chatbots could amplify conflict instead of helping de-escalate it.