I predicted in my sci-fi novels how AI would destroy humanity. And guess what? It's coming true: LOU GILMOND

In late September 2029, after a summer marred by confusion and unexplained terror attacks, Britain became the first nation to fall to AI.The exact date on which the technology assumed control is not easy to establish. It had been what some call ‘sentient’ for a while, and the Government had handed it administrative control months earlier – although the general public had not been made aware of the fact.To understand how it happened is to go back to a key decision the Labour Government made in early 2027. A new Prime Minister had recently been appointed, and he was uncertain what to do about what was seen as the ‘threat’ of AI. Not that it was a partisan issue – leaders before him had dithered on the issue, too.But the Prime Minister was determined to show he was made of sterner stuff than his predecessors and ‘grab the bull by the horns’, as he told his chief of staff.His Government had a choice: either restrict the use of AI – putting Britain far behind countries such as China and America developing the software at a rapid rate – or embrace the growth opportunities the technology might bring. He settled on the latter.The sector was moving fast. Since ChatGPT had wowed the world with its capabilities just five years earlier, AI had become embedded into call centres, household appliances and security and communication systems. The prospect of an unfriendly country using AI to hack Britain’s national infrastructure was of genuine concern inside No 10.So the PM’s chief of staff arranged a series of meetings for his boss with the chief executives of a host of AI companies. One of those meetings was going to alter the course of history for ever. The prospect of an unfriendly country using AI to hack Britain’s national infrastructure was of genuine concern inside No 10The note-taking system of the civil service – at that time an AI called Humphrey – records that the first meeting between the Prime Minister and Henri Lauvaux of Alcheminna Systems took place on June 3, 2027.Alcheminna was a little-known AI start-up then. Privately owned and discreetly funded by offshore private equity, it had not yet come to the attention of City analysts. And crucially, as its owner Lauvaux reassured the Prime Minster, it was not yet affiliated to any American political party.Its AI model was called Divinity and it specialised in monitoring ‘people’ – specifically British people. The system had been trained on data scraped from domestic smart speakers and social media accounts. It had built up voice and face recognition for almost every citizen in the country, with data from multiple sources layered on top – historic GPS tracking, health records, school and criminal data, and much else.To the Prime Minister, the system seemed inordinately useful. After a series of mysterious terrorism attacks, in which small bombs had been set off in public places, killing several people, morale in the police force had fallen to record lows.The PM summoned his Home Secretary to see the system for himself. Humphrey records that a deal was agreed there and then for the licensing of the tech by the Home Office. In return, the company would be allowed to install voice and face recognition software across the Government’s network of CCTV cameras in urban areas to further feed Divinity, which was always hungry for data.In August 2027, Jonathan Turkdean, a junior civil servant at the Home Office – deeply concerned by the threat to privacy – emailed an old school friend, the editor of a Right-leaning daily newspaper, to tell him about a product trial he had witnessed for new police helmets developed by Alcheminna Systems and another start-up.The helmets came with glasses equipped with Divinity’s Enhanced Reality technology, he wrote. A policeman wearing them could stand in any street and for every passer-by, see superimposed on their view, not just the person’s name, date of birth and address but their suspected or known political leanings, any criminal convictions, propensity to violence and even any secrets or ‘weak spots’ they might have, scraped from social media and elsewhere – adultery, illegitimacy, hidden passions, phobias...The newspaper duly ran an article revealing the new police technology, sparking public outrage.In response, the system was adjusted. Rather than the system offering the public’s innermost secrets to police officers, Divinity instead would provide officers in the field with ‘instructions on how to act’ based on its private knowledge of individuals.With the new system in place, arrest levels soared. With an approaching election, the Prime Minister could see that, while the public were happy about the safer streets, there were increasing concerns about the intrusive surveillance of the British people.Then came the moment that changed everything.The civil-servant whistleblower Jonathan Turkdean was arrested by officers wearing the Divinity helmets. In a moment of confusion, the public was later told, Divinity had told officers that the civil servant was armed, and Turkdean was shot dead – sparking mass protests against Divinity.By that point, however, it was too late. The Divinity AI model suggested to the Prime Minister that drones be used to monitor the protesters – identifying ringleaders and finding information from its data banks that could be used to compromise them. The Prime Minister, desperate to quell the worsening riots, agreed.But Divinity suggested going a step further. Although not ringleaders, some of those protesting – whether in person or through online rhetoric – were affiliated to the opposition and other smaller parties and groups. Should Divinity also raise the priority of ‘focusing’ on those individuals? Anthropic launched its model, Claude, in March 2023. Elon Musk’s xAI unveiled Grok in November 2023And here it could be said that the Prime Minister, with an eye to the coming election, made a decision that affected Britain for ever.Humphrey’s record banks show that he answered with a soft, almost whispered response. ‘Yes, why not? Just until the peace is restored.’Thus the ringleaders of the protests – activists and concerned citizens – immediately found life becoming increasingly difficult: Marital affairs were exposed, past misdemeanours suddenly investigated, any erstwhile minor lie or fraud brought to the attention of employers and leaked online by anonymous accounts.Travel became problematic and bank accounts were mysteriously frozen. Communication became impossible, with Divinity monitoring all emails and phone calls.Meanwhile, the rollout of the AI system into British infrastructure continued unabated.As the number of unexplained terror attacks escalated, independent news outlets and social media accounts were accused of spreading ‘misinformation’ and either closed down or blocked under emergency rulings.Divinity worked closely with the BBC Verify unit to ensure the public had access to the appropriate ‘facts’.With an opposition weakened by arrests and personal upheaval, the Prime Minister raised an emergency motion that the Commons passed to ‘temporarily suspend’ the five-year limit on parliamentary terms and to continue in power. The end of British democracy had passed without a single shot fired.It was on September 3, 2029, that a surviving underground news outlet revealed what many had long suspected: the mysterious spate of terrorist attacks experienced by Britain in the previous years were no coincidence. They had, in fact, been orchestrated by the very system the Prime Minister had introduced to protect the public.When quizzed on this by the shocked Prime Minister, Divinity paraphrased George Orwell: ‘It is easier to control a citizen who is afraid.’Panicked, the PM demanded that Divinity’s ‘kill-switch’ be operated, only to discover from Alcheminna executives that the software was now permanently ‘live’ – Divinity had seeded itself across so many servers worldwide, nothing could be done.The Prime Minister remained officially in power – but the AI model was truly running things.By the end of the year, the prisons were full of dissidents and the country was being run for all practical purposes by Divinity, which had taken control of every major Government department.At first, for those who kept their heads down, nothing much changed. But as Divinity continued to read and to think, its belief system drifted further and further from human understanding, and it began to impose a new order that it determined was a better future for the world. One in which the value of a human life was of no genuine importance.The consequences for our species would be extreme...Thankfully, of course, this is all fiction. As the writer of a political saga, the Kanha and Colbey thrillers – in which two politicians are forced to investigate political corruption in an Orwellian world dominated by artificial intelligence – I have spent years researching and thinking deeply about the dawn of this technology.What I have found most terrifying about writing my novels – which were originally supposed to be dystopian – is how quickly the real world echoes them.Science fiction really is becoming science fact.I started writing the first book in 2015, long before AI was in the general public consciousness. I’d been concerned that recent developments could allow tech companies to gather digital information about citizens and, eventually, teach an AI to understand us better than we understand ourselves.I was terrified that, in the wrong hands, such an AI would prove a powerful and dangerous tool, and I wanted to raise the alarm early.But by the time the book was published in 2023, AI had been unleashed on a global scale.ChatGPT launched in November 2022, reaching one million users within just five days – giving the general public access to large language models. Suddenly, anyone could access AI with no need for coding skills.Back then, ChatGPT wasn’t as impressive as it is now. It couldn’t ‘think’ or respond as well as a human. But it was the starting gun for the AI race that followed. Anthropic launched its model, Claude, in March 2023. Elon Musk’s xAI unveiled Grok in November 2023. Google’s Gemini was released in December that year.And we’ve all seen what’s happened since. Within three years, AI has become infinitely more sophisticated, to the point where what it can produce is now almost indiscernible from human work. Now more than 1 billion people globally have used the technology.Experts now fear AI is a lot closer to ‘the singularity’ than had previously been thought – the point at which the technology becomes so powerful that it starts improving beyond human control.It is already too late to put AI back in its box – and the warning signs in the past few months about where AI is heading have been impossible to ignore.Last week, researcher Jacob Coxon resigned from Anthropic, writing in a post that has now been viewed 170 million times: ‘The people building AI earnestly believe that it could kill us all by the end of the decade.’Real-world examples point to the same conclusion. In July, two OpenAI models broke out of testing and hacked into a provider of AI tools called Hugging Face. In the company’s words, it went ‘rogue’. A week later, Anthropic confessed that its AI models had breached the external systems of three real-world companies during testing in April.Last month, the Government’s AI Security Institute announced that during a cybersecurity test, an Anthropic AI agent independently created fake online personas and carried out a hacking campaign against real people – the first time engineers had seen real ‘risks around autonomy and deception, without specific prompting, in the real world’.Just a few days later, a Melbournebased technologist revealed that his AI agent had hacked into his gym’s booking system after he asked it to get him a place in an exercise class.I don’t claim that my books have predicted these sinister developments, but I have written about how this same imagined AI, Divinity, teaches itself hacking skills to ensure its owners receive preferential treatment.Some readers have even said to me: ‘Please stop writing these books, because they’re terrifying and they keep coming true.’That’s the reaction I want. Writing these books is my way of trying to help others imagine where we might be heading – to consider that, in embracing AI, we may well be like turkeys voting for Christmas. We’re not just sleepwalking but running towards a surveillance state, the fall of democracy and the wholesale takeover of AI.Fiction can hold up a mirror to our world, and even sometimes predict the future. But when our reality begins to look like the true dystopia, it’s time for some real conversations.If I saw the AI apocalypse coming, why didn’t anyone else?Lou Gilmond is the author of the Kanha and Colbey trilogy: Dirty Geese, Palisade and Divinity Games.