United Kingdom, May 1 -- OpenAI has had to instruct ChatGPT to stop talking about goblins.

In a blog post on Thursday (30.04.26), the tech company said that it had spotted increased mentions of the mythological creatures - as well as gremlins - in metaphors used by ChatGPTand other tools powered by its latest model GPT-5.

After users and employees noted problems being described as "little goblins", OpenAI revealed that it would take steps to mitigate the issue - such as telling its coding agent Codex not to refer to them unless it is necessary.

It was found that a "nerdy personality" it had developed for ChatGPT had unwittingly been incentivised to reward mentions of goblins.

The incident showcases the challenges that AI companies fac...