Why ChatGPT creators were forced to stop chatbots from mentioning goblins

In recent months, users of ChatGPT began experiencing a peculiar issue with their responses. For some reason, the replies to their queries made frequent and seemingly unnecessary references to goblins, the mischievous creatures of European folklore.

Was it a hangover from Oxford’s 2022 Word of the Year being the term ‘goblin mode’? Or had the chatbot perhaps discovered that the creatures are actually real, with their existence covered up by governments around the world?

After a period of silence from OpenAI, the truth has finally emerged.

In a statement entitled 'Where the goblins came from', the company said it all stemmed from its creation of a ‘nerdy personality’ for the then-new ChatGPT 5.1, which unintentionally prioritised references to the creatures, and in its words, “from there the goblins spread”. 

The company said it first became aware of the problem in November last year, with the unwanted mentions continuing to grow.

By the time staff looked into it, references to goblins had risen by 175% since the launch of ChatGPT 5.1, with the instances of the term ‘gremlins’ also seeing a baffling spike in frequency.

However, the issue was deemed relatively minor at a time when OpenAI was facing more significant challenges, such as a lawsuit by co-founder-turned-bitter-rival Elon Musk.

As the company admitted, this oversight came back to bite it in March this year when ChatGPT 5.4 was rolled out.

Facing an influx of goblin references, an internal investigation was held, which finally found the cause of the problem. Since then, OpenAI says it's resolved the matter, leaving users free to consult ChatGPT without navigating unwelcome mentions of goblins or gremlins. 

At a time when fears about our reliance on Artificial Intelligence are increasing, the case is a sobering, if humorous, example of how a small mistake can rapidly multiply with dramatic consequences.

Although humanity was able to avoid any goblin-related damage this time, we might not be so lucky in the future.