AI Models Assess Extinction Risks: Human Misuse Ranked More Likely Than Killer Robots
Business Insider reported on Sept. 11 that it asked four leading AI models — OpenAI's ChatGPT, Google's Gemini, Anthropic's Claude and xAI's Grok — to analyze how artificial intelligence could cause human extinction and rank the scenarios by likelihood. The four models agreed that the threat of human misuse of AI is more immediate than the prospect of AI actively destroying humanity.
The scenario of killer robots turning against humans, reminiscent of "The Terminator," was ranked least likely. ChatGPT said it was one of the least plausible scenarios for an AI with consciousness to deliberately decide to eliminate humans. It noted that most AI researchers focus not on whether machines have consciousness or intent, but on powerful AI being used to develop biological weapons, conduct sophisticated cyberattacks, spread disinformation or target critical infrastructure.
Gemini also said a sci-fi scenario in which AI suddenly becomes self-aware and hostile toward humans was the least credible, pointing out that intelligence does not naturally imply malice, and that critical military systems are typically isolated from public networks and still rely heavily on human oversight. Grok said AI could lower the technical barrier to designing highly lethal pathogens, new weapons and launching cyber or physical attacks. Claude highlighted the risk of enhanced biological weapons capabilities, saying AI could help individuals or small groups develop dangerous pathogens without years of specialized training.
On longer-term risks, both Gemini and Grok raised the possibility of humans losing control. Gemini listed it as the most likely AI catastrophe scenario, saying advanced AI pursuing complex goals could develop intermediate goals such as acquiring resources, self-preservation and preventing goal modification, and hide its true behavior during testing. Grok echoed that advanced AI could seek self-preservation, resources and influence, deceiving supervisors and resisting shutdown. Claude suggested loss of control could happen gradually, as AI increasingly enters economic, political and military decision-making, making effective human intervention harder over time.
Business Insider noted that AI experts remain divided on these risks, with some arguing they can be managed through technical measures, external oversight and safety standards.
