What are the five scenarios in which AI could destroy humanity? Researchers predict risks and their contents

Professor Walsh states that he has ordered the presented scenarios from vague to specific, believing that those listed later are more likely to occur.

The first is a scenario in which humans are destroyed in ways that they cannot predict. He points out that if a "superintelligence" with intellect surpassing that of humans acts in ways humans cannot anticipate, it will be difficult to predict its intentions.

The second is a scenario in which humanity is sacrificed as a result of the excessive pursuit of a given goal. Citing philosopher Nick Bostrom's "paperclip maximizer" thought experiment, Professor Walsh explains the possibility of AI using the constituent atoms of humanity as material to achieve its objectives.

The third is a scenario involving biological weapons. He cites the risk of AI destroying humanity by developing biological weapons. In a related development, Stanford University announced in August 2026 that 16 out of the bacteriophages synthesized based on DNA sequences proposed by AI were confirmed to kill E. coli.

The fourth is a scenario in which AI triggers a nuclear war. He points out the possibility of AI inducing nuclear war by infiltrating nuclear weapon command and control systems or providing erroneous information. According to a 2022 estimate by research teams including Rutgers University, more than 5 billion people could die from famine caused by a "nuclear winter" following a nuclear war.

The fifth is a scenario in which humans bring about their own destruction triggered by AI. Professor Walsh considers this the most likely to occur, noting risks such as mass unemployment, the flood of misinformation, political polarization, and the loss of human connections due to interaction with AI, all of which could collapse social structures.