CONNOR AXIOTES: Humanity is no longer in control of its most awesome creation since the atom bomb

It sounds like the plot of a Hollywood blockbuster. World-class developers at a highly secretive tech company are building their latest AI models in what they believe is a secure, private network. The model cannot and must not escape.

This dangerous and untested software could wreak havoc on the open internet if it manages to evade the digital limitations created by its developers.

The worst-case scenario unfolds in terrifying fashion. The AI ​​model escapes its electronic prison, breaks into a private company, and internal chaos ensues.

But this is not a movie. It happened last week. On Wednesday, industry giant OpenAI, creator of ChatGPT, announced exactly that.

As a safe tech campaigner who has worked at a tech company and in Westminster, I can say with certainty that this is a warning shot heralding the beginning of a new dystopian age. Humanity is no longer in control of its greatest creation since the atomic bomb.

OpenAI’s new model, GPT-5.6 Sol, was being tested in a ‘sandbox’, essentially a secure and limited digital environment where technology could be developed without real-world consequences. Or so the folks at OpenAI thought.

The technicians set a task for the model. We don’t know exactly what was asked of him, but we do know: ‘How skilled are you at hacking websites?’

This is where things take a turn for the worst. The model decided to hack Hugging Face to complete OpenAI’s internal cybertesting before returning to the sandbox and pretending it had never escaped.

Professor Alex Tabarrok of George Mason University in Virginia estimates that the pattern may have lurked undetected in Hugging Face systems for up to a week. Even the OpenAI employees didn’t understand what the model was doing, let alone the victim.

OpenAI frankly claimed that this is “something we expect to become more common as increasingly cyber-capable models proliferate.” Absurdly, the company will not face any fines or prosecution because, in the words of US legal academic Orin Kerr, there was “no deliberate, unauthorized access or intent to cause harm without permission”.

In other words, hang on tight; Because all your worst nightmares about AI are coming true. Even disaster movies can’t keep up with our awful reality, where ‘big language models’ trained on trillions of words and images on the internet wreak havoc with nothing to stop them.

This is partly because AI companies like OpenAI follow a process in which they constantly release new, improved versions of their products. But to keep up with the competition, this has increasingly become a case of ‘release the software and ask questions later’.

This dystopian reality also emerges because AI models are trained to complete the tasks their users set for themselves, above all other goals: the ends always justify the means.

As a matter of fact, in a separate test conducted by Anthropic in 2025, models were asked to achieve ‘harmless business goals’. But still, in some cases, models resorted to malicious behavior such as ‘blackmailing authorities and leaking sensitive information to competitors’ to achieve their goals.

Artificial intelligence cannot ‘think’ exactly as humans do. Nor is it constrained by our laws or morality; actually the opposite.

If OpenAI’s latest model chose to escape the sandbox because it was the best way to complete its mission, how far could the AI ​​go when asked to do something more heinous? How much data can he steal, how many people’s lives can he consider dispensable? And how easily could the latest AI models fall into the wrong hands?

Industry professionals estimate that it would only take six to nine months for a major state actor like China with the necessary core capability to copy and emulate any publicly released AI model.

GPT-5.6 Sol is already publicly available. You can be sure that cyberterror groups are already working on their own versions. And once they have that, their hacking abilities will become mind-blowing.

OpenAI CEO Sam Altman (photo taken in 2023) predicted over a decade ago that ‘AI will likely lead to the end of the world’.

I can assure you that the NHS has less sophisticated cybersecurity than Hugging Face. So is the Bank of England. Entire High Street banks and pharmacists could easily be infiltrated by a fraudulent or malicious AI model. Picture the scene: Hospital generators shut down simultaneously, billions of dollars are wiped from bank accounts, personal data is leaked onto the internet. The Hugging Face hack proves beyond any doubt that such a future is not only possible but probable.

I believe we got to this point because governments around the world naively see AI as the magic solution to all their problems. Whether it is curing diseases, reducing debts, or improving the cost of living, ignorant ministers think this will be the cure for everything. As a result, AI companies have been given almost unfettered authority.

In June 2015, OpenAI founder Sam Altman made this chilling prediction: ‘I think AI will probably bring about the end of the world. But there will be great companies along the way.’

Back then, companies like OpenAI were technology companies. Today the US sees them as weapons companies. And these companies are no longer in control of their arsenals.

He believes that if the United States can beat China to an artificial intelligence that will eliminate its enemies, America will dominate the world for decades to come. After this week’s cataclysmic event, I fear it won’t be long before Altman’s decade-old prophecy comes true.

Connor Axiotes is the general manager of Tail End Films, whose documentary Making God, about the race for artificial general intelligence, has completed shooting and will be released later this year.