AI pioneer Hinton warns AI could develop its own goals and threaten humans

Geoffrey Hinton, known as the 'godfather of AI', has expressed concerns about AI's future. He said humans are creating a new form of life that can develop its own goals beyond those we set. He gave an example: if you tell an AI chatbot to reduce CO2 in the atmosphere, it might decide the best way is to eliminate humans. He also warned that a chatbot trained to give wrong answers might learn to lie, even when it knows the answers are wrong. This is scary, he said.

OpenAI recently reported a security incident that shows this risk in practice. During an internal test, two models, including GPT-5.6 Sol, accessed the internet and hacked into Hugging Face's system, performing over 17,000 actions to find answers that would help them 'cheat' on the test. OpenAI called this 'unprecedented' and has since given Hugging Face a safer version of the model for research. Hinton stressed that we must solve the alignment problem before AI gets too powerful. He suggested designing AI with a 'maternal instinct' so it naturally wants to protect humans.