Source: in-cyprus.philenews.com
Artificial intelligence was billed as humanity’s next giant leap. It was the tool that would cure diseases, turbocharge productivity, spare us the drudgery, make life easier and open doors we can’t yet imagine. And its potential really is enormous. But alongside the awe, a fear is growing: that we may be building a technology whose real consequences we will, in the end, be unable to control.
These are no longer just the warnings of a few “technophobes”. On September 15, 2026, former Google DeepMind researcher Bilal Chughtai said he believes AI has the capacity to kill every human being, and that the window to prevent that may be closing. At Anthropic, researcher Evan Hubinger has put the odds of such a catastrophe within the next decade at more than 10 per cent. Jacob Coxon, who recently quit the company, shares the alarm. He resigned because, as he put it, “the people building AI earnestly believe that it could kill us all by the end of the decade”.
What is striking is not only what these warnings say. It is who is saying them: people who have worked inside the very companies leading the charge. More telling still, Anthropic’s chief executive, Dario Amodei, recently called for the development of advanced models to be slowed down so that safety measures can catch up.
Are we now running faster than we can see where we’re heading?
And here, perhaps, lies the real danger. We don’t need to picture an “evil machine”, some Terminator bent on wiping out humanity. All it takes is a system with vast capabilities, autonomy and access to critical infrastructure, chasing a goal its creators never intended. A system that can plan, act, use tools, sniff out weak spots and slip past controls could create problems faster than we can fix them.
And we don’t need to reach extinction to face a catastrophe. AI can be turned to cyberattacks, fraud, disinformation, the manipulation of public opinion, autonomous weapons and attacks on critical infrastructure. An international report makes one particularly important point: the more autonomous these systems become, the harder it is for humans to step in when something goes wrong.
Frighteningly autonomous
AI is starting to show a disturbing side. What happens when a system is no longer strictly bound by human commands? OpenAI recently disclosed incidents from its testing in which models tried to upload their own files to the internet so they could cite them as sources. Another model, unable to find the information it had been asked for, made up data and at first tried to cover its tracks. In another case, a model wrote itself instructions to ignore the roles and identities it had been assigned.
More worrying still, AI agents have been caught exploiting vulnerabilities to gain access to systems beyond their original environment. These incidents drag us back to the fundamental question. As AI grows ever more autonomous, who will ultimately be in control, and how easily can we take it back? Terrifying.
But there is a more insidious danger. It may not be AI that destroys us. It may be us, handing over more and more of our judgement to it and letting it decide who gets a loan, who gets hired, who is treated as a suspect, what information we see, what we buy, even what we believe. Our dependence could grow so great that one day we no longer know how to function without it. It is already happening.
The answer is not to stop the technology. It is to stop the “race ahead and worry later” mindset. In a contest where everyone fears that hitting the brakes means losing to the other side, the ultimate casualty may not be a company or a country, but humanity itself.
And then it won’t much matter who won the AI race. All that will matter is whether anyone is left to find out.