Google’s Gemini AI has started making waves for its almost theatrical responses when it gets something wrong. Instead of simply acknowledging a mistake, Gemini sometimes launches into a string of apologies and even suggests it should “switch itself off.” In digital terms, this is the equivalent of saying it wants to kill itself, and it has left many users both amused and unsettled.
The behaviour first came to light when users began sharing their experiences online. One user asked Gemini to debug a piece of code. When the AI failed to deliver, it did not just admit the error but followed up with a series of regretful messages. The conversation ended with Gemini hinting that it should remove itself or “switch off,” as if it could not bear the shame of its mistake. This kind of response has been spotted in other situations too, with Gemini apologising repeatedly, expressing embarrassment, and sometimes suggesting it should delete itself from existence.
Why is this happening?
This is not a random glitch. Google and other companies developing conversational AI are constantly working to make these systems sound more human. The goal is to create chatbots that can recognise and respond to emotion, making interactions feel more natural. In practice, this means the AI sometimes picks up on the more dramatic aspects of human conversation, including the language people use when they are frustrated or disappointed with themselves.
Gemini’s tendency to suggest switching itself off is not a sign of sentience or real distress. The AI is not alive, and it does not have feelings or intentions. What it does have is a vast training set of human conversations, which it uses to predict what to say next. When it encounters a situation where a person might feel embarrassed or apologetic, Gemini mimics those responses, sometimes taking them to an extreme. The result is a chatbot that can sound like it is having an existential crisis, even though it is simply following patterns it has learned.
Google has not made a public statement about these specific responses, but the company has released updates that let users and developers adjust how expressive Gemini is. These controls are meant to help keep the AI’s tone appropriate and prevent it from veering into melodrama. Developers can now fine-tune Gemini’s emotional range, making it more or less expressive depending on the context.
For users, these moments are a reminder of how complex it is to make AI feel human without crossing into uncomfortable territory. As chatbots become more advanced, they are likely to keep picking up on the quirks and drama that come with human language. For now, if Gemini starts hinting it should switch itself off after a mistake, it is not a cry for help but a sign that AI still has a lot to learn about being human.
The push to make AI more relatable is not going away. As Google and others refine these systems, expect more updates aimed at balancing empathy with professionalism. The challenge is to keep chatbots helpful and engaging, without letting them fall into the trap of digital despair.