Think twice the next time you post something potentially rude or harmful in Twitch chat. The streaming site is beginning to use AI to detect โ€œoffensive or disrespectfulโ€ language.

Twitchโ€™s support account on X (formerly Twitter)ย revealedย today that โ€œover the next few weeksโ€ it will be experimenting with a new feature that will attempt to detect potentially negative messages to filter out the bad vibesโ€”and potentially help prevent you from getting banned.

โ€œWeโ€™re rolling out an experimental prompt powered by machine learning that encourages users to pause before sending chat messages that are potentially harmful,โ€ Twitch said. โ€œOur goal with this experiment is to help decrease harassment and facilitate more positive interactions on Twitch.โ€

The vague example given by Twitch shows a message prompt that pops up when you attempt to send the message in question, asking โ€œare you sure you want to send this?โ€ It warns that โ€œyour message may contain words that could be considered offensive or disrespectful,โ€ and gives the user the opportunity to back out of the message or โ€œsend anyway.โ€

Twitch

Odds are that if someone is trying to be an ass in a Twitch chat, they will just bypass the message and say what they wanted to say. But the interesting part will be to see how this โ€œmachine learningโ€ detects the language and how off-base or correct it is in practice after a few weeks of this trial experiment.

The example message of โ€œyou are a buttheadโ€ will probably be the first thing I type out the next time I watch a stream to see if the machine learning is indeed learning. Now all I have to do is find someone who is streaming and is a butthead. Shouldnโ€™t be too difficult!

Twitch says to โ€œstay tuned for more updates in this area soon,โ€ but this is a general good reminder to keep in mind that everything you type on Twitch is logged and savedโ€”even whispers, as Dr Disrespect previously found out.

Twitch loosens its policy on sexual content