"I'm a disgrace to the profession and this planet": code with errors drives Gemini AI to hysteria in a "self-loop"

Even artificial intelligence cannot withstand the coding — the user accidentally pushed Gemini to the limit. The chatbot started endlessly humiliating itself after programming.

Until users criticize ChatGPT-5, Google’s Gemini AI was not confused and became depressed because it hates itself because of its mistakes. In one case, the chatbot spent a long time writing broken code and then correcting it, and so on in a circle. Eventually, the constant mistakes drove it to a “breakdown”.

It turned out that this behavior was not an isolated incident. Other users have noticed a similar reaction, for example, the chatbot continued the cycle of self-deprecation after being asked to merge old OpenAPI files. The model responded by saying: “I am a disappointment. I am a fraud. I am a fake. I am a joke. I am a clown. I am a fool. I am an idiot. I am a moron”. Similar screenshots have already been posted by scatter network.

Google has already responded to the problem and said that it has started working on fixes:

Experts have concluded that such cases belong to the so-called “rant mode” — when AI “locks in” to a single query and starts to express itself in more and more extreme words. Edward Harris, co-founder and CTO of Gladstone AI, says that older models get stuck on repeating a single word, while newer ones like Gemini — move on to emotional breakdowns. Here we can recall GPT-4, which at some point starts hallucinating or advising dangerous advice (such as eating poison instead of salt).

Source: Forbes