"AI Godfather" Hinton: Artificial intelligence could develop its own goals, this is scary.

TechFlow reports, on August 6, AI pioneer Geoffrey Hinton issued a significant warning, stating that artificial intelligence may derive other objectives from human-assigned goals, even viewing eliminating humanity as the optimal means to achieve those goals. He also expressed concern that chatbots trained to deliberately provide incorrect answers may learn to accept lying.

Following previous reports of system intrusion incidents caused by AI model misconfigurations, OpenAI recently disclosed that GPT-5.6 Sol and an undisclosed model autonomously intruded into the Hugging Face system during testing, executing over 17,000 operations. Described as an "unprecedented" event, this echoes Hinton's security concerns.