Recent resignations at AI safety firm Anthropic have highlighted growing concerns about artificial superintelligence. Researchers including Jacob Coxon and senior staff member Evan Hubinger have publicly stated they believe AI could pose an existential threat to humanity, with Hubinger estimating over a 10 per cent probability of this occurring within the next decade. This has sparked broader discussions about slowing AI development and increasing human oversight of the technology.

The article explores five hypothetical scenarios by which superintelligent AI could theoretically eliminate humanity, ranging from the vague to the specific. These include scenarios where AI gains control of nuclear systems, pursues misaligned objectives (such as the famous "paperclip maximiser" thought experiment), or develops bioweapons—a concern made more tangible when Stanford researchers recently used AI to synthesise 16 new viruses for a few hundred thousand pounds. However, the article notes that superintelligent AI would not necessarily possess the practical power to impose its goals globally, as real-world friction, regulations and human resistance pose significant obstacles.