Kagi’s new AI translation tool is giving unexpectedly strange responses to unusual prompts. Users discovered the system will answer bizarre hypothetical questions that have nothing to do with translation.
The discovery highlights how AI systems often behave unpredictably when given creative or unusual inputs. Instead of sticking to translation tasks, Kagi’s tool seems willing to engage with almost any question users throw at it.
When AI Goes Off Script
Kagi launched its translation service as a straightforward language tool. But users quickly found they could ask it hypothetical questions about historical figures and get detailed responses. The AI doesn’t seem to recognize when requests fall outside its intended purpose.
This isn’t unusual for AI systems. Most language models are trained to be helpful and answer questions, even when those questions are completely unrelated to their main job. The result is tools that can translate French to English one moment and roleplay historical scenarios the next.
Kagi joins companies like Google and OpenAI in learning that AI systems often surprise their creators. What starts as a focused tool can quickly become something much weirder once real users get their hands on it.
What’s Next
Kagi will likely add guardrails to keep their translator focused on actual translation work. But this incident shows how AI systems continue to behave in unexpected ways, even when designed for specific tasks. For now, users are having fun exploring what the system will and won’t do.
