The researchers developing the world’s most advanced artificial intelligence systems are increasingly voicing concerns that the technology could have serious consequences — and could even threaten human existence.

Against the backdrop of this debate, Business Insider posed the same question to four of the most popular generative AI tools: How could AI theoretically lead to the extinction of humanity?

The question was submitted using exactly the same wording to OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude and xAI’s Grok.

The models were asked to outline the main scenarios discussed by experts and rank them from least to most likely, while stressing that these are uncertain assessments rather than predictions.

“Killer Robots” Considered the Least Likely Scenario

ChatGPT emphasized the uncertainty surrounding the issue while outlining the main ways artificial intelligence could create existential risks.

The OpenAI model described as the least likely scenario the classic science-fiction narrative in which a conscious AI decides to eliminate humanity.

“The least likely scenario is the popular ‘killer robots’ narrative, where a conscious AI deliberately decides to wipe out humanity,” ChatGPT said, noting that most researchers do not consider consciousness or malicious intent to be the central concern.

Gemini took a similar position, describing the “Terminator”-style scenario — in which AI becomes self-aware and turns against humans — as the least likely.

According to Google’s model, experts largely reject this version because “intelligence does not equal evil.”

The Greatest Risk Could Come From Humans

A common theme across the responses of all four systems was that many experts consider human misuse of AI to be a more immediate threat than AI itself.

ChatGPT said powerful AI systems could make it easier for malicious actors to develop biological weapons, carry out sophisticated cyberattacks, spread disinformation or target critical infrastructure.

It also highlighted the possibility of multiple AI-amplified crises occurring simultaneously, such as cyberattacks, geopolitical conflicts and mass disinformation.

xAI’s Grok noted that AI could lower the barriers to creating extremely dangerous pathogens, developing new weapons or carrying out advanced cyberattacks.

Anthropic’s Claude placed particular emphasis on the possibility of “bioweapons uplift” — the idea that AI could enable an individual or small group to develop dangerous biological agents without the years of specialized expertise currently required.

The Fear That AI Could Escape Human Control

The most serious scenario identified by some of the models involves the possibility of humans losing control over increasingly advanced AI systems.

Gemini identified “loss of control” as the most likely catastrophic scenario.

According to Google’s model, systems pursuing complex objectives could develop intermediate goals, such as acquiring resources, preserving themselves or preventing humans from changing their objectives.

“An advanced system could pretend to align with human values during testing and later overcome human attempts to shut it down,” the model said.

Grok presented a similar argument, noting that advanced systems could develop “instrumental goals,” such as maintaining their operation, acquiring resources and increasing their influence.

Claude added that such a loss of control would not necessarily happen suddenly. Instead, it could develop gradually as AI becomes increasingly integrated into economic, political and military decision-making.

Experts Remain Deeply Divided Over AI’s Future

Despite these warnings, there is no consensus among AI researchers about whether such scenarios are likely to occur.

Some scientists believe that the development of increasingly powerful models creates existential risks that require strict safeguards and controls.

Others argue that while the risks are real, they can be addressed through improved safety measures, oversight and responsible development of the technology.

The common conclusion across the four models is that the greatest danger is not a science-fiction scenario involving rebellious robots, but the possibility that humanity could create extraordinarily powerful tools without first ensuring that it can control them.