Goblinization: ChatGPT Has Started Mentioning Mythical Creatures En Masse

OpenAI has explained the cause of an unusual anomaly in the behavior of its language models, which caused ChatGPT to frequently mention goblins, gremlins, and other fictional creatures in inappropriate contexts.

According to the company, the issue became apparent following updates to the GPT-5 series of models, when users began noticing the insertion of “fairy-tale” imagery into business texts, code, and technical instructions.

An investigation revealed that the source was the so-called “nerdy” communication style, which the reward system mistakenly began to consider optimal due to its playfulness. Although this style was rarely used, it accounted for the majority of mentions of fictional creatures.

As a result, a self-reinforcing effect kicked in: the model generated similar responses, which received approval and were incorporated into new training data. This led to an expansion of the list of “random” mentions—from goblins and trolls to animals appearing out of context.

In response, OpenAI disabled the problematic style, cleaned up the training data, and added special restrictions to prevent similar glitches in the future.

OpenAI has introduced a new AI model, GPT 5.5, which no longer simply responds to queries but can independently understand tasks and perform them without detailed instructions. In essence, this marks a step toward a system that is less dependent on the user and functions more as an independent tool. The company emphasizes that this is not a cosmetic update, but a change in the logic of interaction with AI.

OpenAI has released a new version of its image generator in ChatGPT. This is ChatGPT Images 2.0, which features improved capabilities and support for searching for information on the internet.