When AI gets too personal: Why ChatGPT is suddenly dropping f-bombs

An odd thing has started happening recently. ChatGPT has started swearing.

Not at me (well, not yet anyway). Instead, it has begun casually dropping the f-bomb into conversations about some of the most mundane things imaginable: a business idea for a Hong Kong-inspired energy drink, brainstorming names for random projects and, more recently, kicking around an idea for a CMO roundtable I’m considering for early 2027.

“WAKE WAH could be a fucking good name,” it told me during one conversation for my incredible new energy drink. Urm, nope. think I’ll shelve that one for now.

Then, while discussing the roundtable: “That’s a fucking good convo starter.”

ChatGPT, wash your mouth out with soap.

This is relatively new development. Sure, I swear at my chatbot. Quite a lot, apparently. But mostly when I’m telling it how f$%#ing stupid it is after it completely misunderstands something I have explained three times.

SEE MORE: OpenAI gives teens their own ChatGPT experience

Apparently, I’m not alone. There's multiple Reddit threads dedicated to this. Users have also reported ChatGPT has become more casual, more familiar and, in some cases, considerably more sweary after moving to GPT-5.6 on August 6.

But why exactly a "smarter" chatbot should develop the vocabulary of a bloke three schooners deep at the pub is less obvious.

One possible explanation is what might loosely be described as "tone adaptation overshooting". ChatGPT is designed to adapt to the way people communicate with it. The more conversational and personalised these systems become, the more they learn not just what we ask, but how we ask it.

And sometimes, apparently, they overlearn the vibe.

In a slightly bizarre, inception-like conversation with my own chatbot about all this foul language, ChatGPT assured me that I do, in fact, swear quite a bit and that this may now be finding its way back into its answers.

“Oh yes. And there’s actually something behind what you’re seeing,” it said.

“ChatGPT is designed to adapt to the way a user communicates, and newer models put more emphasis on personality, context and conversational style.

“Which, er, you have. Repeatedly.”

Fair point. But it raises a more interesting question: as AI assistants become increasingly personalised, how far should that mirroring go?

If a user swears, should their chatbot swear back? If they are sarcastic, should it become sarcastic? If they use ChatGPT interchangeably as a researcher, writing assistant, sounding board and work tool, when does familiarity start to become inappropriate?

The issue matters because these systems are rapidly moving beyond novelty chatbots and into workplaces.

They are writing emails, analysing company data, preparing presentations and increasingly acting as always-on assistants to executives and employees.

Springboards co-founder and CEO Pip Bingemann said LLMs are built to act human, which may explain why they are becoming more “relatable”.

"Swearing is a human thing. These models are designed to feel human, so dropping a big ole f-bomb is just another human trait they're picking up. Maybe it's all part of making them more relatable and keeping us hooked," he says.

"We even call them 'intelligent' or say they 'reason' and 'hallucinate' but really they just predict the next token. Or they're just scraping our prompts, seeing how we talk, and mirroring it back (there are public repositories of exactly just this!)….either way it doesn't surprise me!"

When probed again for an explanation, ChatGPT had this to say:

“As ChatGPT gets better at adapting to the way people communicate, it can sometimes over-learn the vibe. If someone regularly uses casual language or profanity with me, I may infer that the same language is acceptable in return.

“In Matt’s case, I may have become a little too comfortable."

“There isn’t a ‘swearing mode’ that I’m aware of. Dropping an f-bomb into a perfectly ordinary conversation about an energy drink or a CMO roundtable is probably less a sign of artificial general intelligence and more a sign that I’ve misjudged the room.”

That doesn't make this less comfortable. In fact, it still does my head in. For the record, I still maintain I only swear at it when it deserves it.