As AI systems become more capable, they introduce risks that cut across the others on this page. Advanced AI could make bioweapons easier to engineer, be used to entrench authoritarian control, or be integrated into systems like nuclear command and control where errors are catastrophic.
The deeper worry is about systems we don’t fully understand. Even current models don’t always act as their developers intend, and that problem gets harder as capabilities grow. If AI systems become more capable than humans across most domains, losing control of them could be irreversible.
Researchers working on AI safety are trying to address these problems, though the field remains small relative to the number of people building the technology.