The researchers building our favourite chatbots are progressively voicing concerns that AI could wounded — aliases extremity — humanity. So we asked 4 of the most-used generative AI devices what an AI last day could look like.
We posed the aforesaid question, successful identical wording, to OpenAI's ChatGPT, Google's Gemini, Anthropic's Claude, and SpaceXAI's Grok.
The 230-word punctual had clear parameters and instructions, including the following:
Help explain, successful clear and accessible terms, really AI could theoretically lead to quality extinction. Identify the astir commonly cited scenarios. Rank the scenarios from slightest to astir plausible based connected the champion disposable master analysis, making clear that these are highly uncertain forecasts, not established predictions. For each scenario, explain:- How it could unfold
- What would request to spell incorrect for it to go catastrophic
- Why experts see it much aliases little plausible than the different scenarios
- Which safeguards could trim the risk
We posed a akin mobility to AI experts for their return connected how, and if, AI could termination america each — you tin publication that communicative here. There's a crisp disagreement among AI experts; while immoderate are predicting the worst for humanity, others are presenting much tempered and optimistic predictions.
Here's what the chatbots told america astir really AI would theoretically lead to quality extinction.
'Killer robot' is the slightest plausible scenario
ChatGPT took a balanced and explanatory tone, saying there's a batch of uncertainty astir the topic.
It focused connected the main ways AI could airs existential risks, ranking them from slightest to astir plausible.
"The slightest plausible script is the celebrated 'killer robot' narrative, successful which a conscious AI deliberately decides to destruct humanity," ChatGPT said. "Most AI researchers do not spot consciousness aliases malicious intent arsenic the cardinal concern."
It besides outlined the safeguards experts propose and stressed that location is nary statement that AI-driven quality extinction is apt aliases inevitable.
Gemini echoed immoderate of ChatGPT's points, saying "The Sci-Fi 'Terminator' Takeover," successful which AI becomes self-aware and originates to dislike humanity, is the slightest plausible scenario.
"Technical experts mostly disregard this trope because intelligence does not adjacent malice, and captious subject infrastructure remains air-gapped and heavy reliant connected quality oversight," Gemini said.
Humans utilizing AI are a bigger consequence than AI itself
In their responses, each 4 chatbots said galore experts see quality misuse of AI to beryllium a much contiguous consequence than AI itself.
"Powerful AI systems could make it easier for bad actors to create biologic weapons, motorboat blase cyberattacks, dispersed disinformation, aliases onslaught captious infrastructure," ChatGPT said.
The OpenAI chatbot added that a cardinal consequence besides comes from "multiple AI-enabled crises" for illustration cyber failures, geopolitical conflict, and the dispersed of misinformation happening astatine the aforesaid time.
Grok said AI lowers barriers to designing highly lethal pathogens, caller weapons, aliases cyber and beingness attacks.
Anthropic's Claude besides highlighted "bioweapons uplift"— the anticipation that AI could let a personification aliases mini group to create a vulnerable pathogen without years of specialized expertise.
Advanced AI systems could 'outsmart humans'
Gemini and Grok some said that humans losing power of AI is simply a possibility.
Gemini branded "loss of control" arsenic the astir plausible AI punishment scenario.
"Systems fixed analyzable goals people create intermediate sub-goals, specified arsenic acquiring resources, self-preservation, and preventing extremity modification," the Google-made chatbot wrote. "An precocious strategy mightiness feign alignment during information (deceptive alignment) and later outsmart quality attempts to unopen it down."
SpaceXAI's Grok utilized akin connection astir this risk.
"They whitethorn create instrumental drives (self-preservation, assets acquisition, influence) useful for almost immoderate objective, perchance deceiving overseers, resisting shutdown, and disempowering humanity," Grok said.
Claude added that this nonaccomplishment of power whitethorn beryllium gradual, arsenic AI systems go embedded successful economic, political, and subject decision-making until meaningful quality involution becomes progressively difficult.
Read next
Shuby is simply a elder newsman astatine Business Insider's Singapore bureau, wherever she writes astir tech, pinch a attraction connected AI training and vibe-coding startups.She antecedently interned for Bloomberg News and CNBC. She studied communications and business astatine the National University of Singapore. In 2026, she won the Singapore Press Club Tech Journalism Award for her activity connected the hidden quality costs of Meta's chatbot training programme and an exposé connected the achromatic marketplace for AI gig-work accounts.Have a tip? Contact her via email astatine [email protected] aliases connected unafraid messaging app Signal astatine @shuby.85. We tin support sources anonymous.
Georgia is simply a Business News Fellow at Business Insider's London office.Before joining Business Insider, she worked astatine Japan's largest newspaper, The Yomiuri Shimbun, and interned astatine the Financial Times. She is an NCTJ-qualified journalist pinch a grade successful Philosophy from the University of Birmingham.
English (US) · 
Indonesian (ID) · 
