Loss of Control Over Advanced AI

Of the plausible extinction-level threats, advanced AI has the steepest growth in capability and the weakest safeguards. Frontier models are improving quickly. They are increasingly autonomous and are being wired into code, finance, infrastructure, and military systems. Meanwhile, the geopolitical rivalry between major powers and the commercial race among companies reward speed over caution. Safety research, interpretability, and international regulation are all lagging behind deployment.

The most likely path is not a sudden "robot uprising." It is a gradual loss of human control. Systems that are more capable than people at strategy, persuasion, cyber operations, and scientific research could pursue goals that are slightly misaligned with human welfare. As institutions come to depend on them, humans could lose the practical ability to switch them off or override them. Misaligned goals, deceptive behavior, or AI-assisted engineering of pathogens or cyberattacks could then cascade faster than any society can respond.

Other risks are real but less likely to end the world outright. Climate change is severe but slow, and it probably degrades civilization without eliminating it. Nuclear war is catastrophic but less probable in any given year. Pandemics and social fragmentation are survivable. Advanced AI could amplify all of these risks, which makes it the central point where they converge.

This is a speculative, low-probability forecast, not a certainty. Its likelihood depends on whether alignment, oversight, and global coordination can keep pace with capability growth.