Overview: the question is serious, but the answer is not a prophecy

“Could AI kill us all?” sounds like the title of a science-fiction film. In 2026 it is also a real question discussed by researchers, AI companies, governments, journalists, and ordinary people trying to understand what rapid progress in artificial intelligence means for the future.

The responsible answer has two parts that must be held together.

First, there are plausible mechanisms by which future AI systems could contribute to catastrophes on an enormous scale. Researchers study loss of control over highly capable autonomous systems, AI-assisted cyber operations, biological and chemical misuse, military escalation, critical-infrastructure failures, deception, concentration of power, and interactions between these hazards. Major AI developers have created safety frameworks specifically because they believe some future capabilities could create severe harm. The 2026 International AI Safety Report devotes substantial attention to these risks and to the uncertainty surrounding them.

Second, plausible does not mean inevitable, imminent, or scientifically quantified. Present-day AI systems are impressive but still brittle. They make mistakes, fail at long-horizon tasks, depend on human-provided infrastructure, and do not independently possess the full combination of capabilities, persistence, access, and strategic reliability that many extinction scenarios require. Experts disagree sharply about how likely future loss-of-control scenarios are. Some consider them important enough to justify aggressive preparation; others think they are implausible or too speculative to deserve the attention they receive.

The goal of this guide is not to choose a tribe. It is to build a map.

A good map asks:

• What can current AI actually do?
• What capabilities would a catastrophic scenario require?
• How could AI become connected to real-world power?
• Where would human decisions still matter?
• Which claims are observations, which are forecasts, and which are stories?
• What safeguards can break a dangerous chain before harm becomes irreversible?

That distinction is the foundation for thinking clearly about AI risk.