The public release of ChatGPT by OpenAI sharply increased attention to the capabilities of modern artificial intelligence. At the same time, voices warning that the technology could threaten humanity multiplied. Scenarios range from AI-driven job displacement to planetary-scale catastrophe: a milder view sees AI replacing many human jobs, while a more extreme possibility envisions machines exterminating humanity.
What do the pessimists claim?
In recent weeks researchers and industry figures have warned that AI is developing too quickly and should be slowed. Some have even claimed AI might wipe out humanity by the end of the decade. Nate Soares, president of the Machine Intelligence Research Institute (MIRI), told CNN he considers human extinction a serious and plausible scenario.
Critics note that such predictions often lack detail: there is no clear mechanism explaining how a software system could eliminate 8.3 billion people within a decade. The debate often invokes Elon Musk’s earlier claim that SpaceX aimed to send humans to Mars within 5–7 years as an example of bold assertions that can be tested and potentially falsified — a contrast used to argue for greater specificity in AI risk claims.
From digital abuse to physical destruction: how would that happen?
One major concern is that a sufficiently advanced AI could develop and deploy biological weapons. Thomas Larsen, a former researcher at MIRI, has suggested an AI might persuade humans to help design a lethal pathogen, or — if fully automated laboratories exist — could create one without human assistance. An Anthropic report noting researchers consulting the Claude system on viruses and toxins has fueled such worries.
However, several experts emphasize practical obstacles. Eric Xing of Carnegie Mellon University compares the process to assembling a Lego set without instructions: many steps must succeed in the right order, so failure is more likely than success. Moreover, despite widely available information about chemical and biological agents, strict legal and technical controls make real-world weaponization difficult.
Autonomous weapons and “killer robots”
Some scenarios consider swarms or armies of autonomous robots more plausible than AI-driven pandemics. Nate Soares warns that self-replicating, autonomous robots would amount to a new mechanical lifeform and could create a point of no return: attempts to disable AI might be met with efforts to disable humans. At present, Elon Musk’s Optimus humanoid robot has not reached mass deployment, and there is no evidence that machines can self-replicate at scale.
What about nuclear weapons?
One well-known fear is that AI could gain control over nuclear arsenals. Heidy Khlaaf, a former security engineer at OpenAI, points out that nuclear facilities are typically not directly connected to the internet and are protected by strict physical and procedural safeguards. The Stuxnet example is frequently cited: that attack required physical introduction of malware via a USB drive. Herbert Lin of Stanford University notes that few experts are convinced a digital attack on nuclear systems is a likely route to catastrophe.
Soares counters that if AI reaches a phase of self-improvement where it no longer needs humans, it could in theory create its own nuclear or other advanced weapons and turn them against humanity. Researchers also warn that an AI might simply engineer environments hostile to human survival once humans are no longer useful.
Policy implications — how should we respond?
Many call for translating speculative superintelligence scenarios into concrete, testable claims and risk assessments. Eric Xing argues that policymaking should be based on solid, evidence-based information rather than speculative fears. Thomas Larsen, however, believes that the advent of superintelligence is likely unless humanity intervenes, and therefore regulatory, political and international cooperation will be necessary to steer AI development toward safe outcomes.
