Plagiarism, attacks and dependencies: the disastrous summer that AI has brought us

August 2026, summer bulletin marking the first year of intensive AI implementation. I still remember the dawn of this technological era when the question was: ‘What can it do?’ What incredible naivety. The good news: it writes those tedious emails for us. The not-so-good news: there are quite a few ways to hack into it, and just as many unintended uses.

Google Earth had introduced a feature that allowed users to select a location on the planet and edit the image. The system was powered by something called Nano Banana 2 (the cutest name ever given to a programme), and it did not generate a landscape-like fantasy. It took Google’s real three-dimensional satellite images and inserted the imaginary event. For example, users could bring Pompeii back to life. The feature remained available for approximately 24 hours; Google removed it the day before yesterday. The usual issues arose: unauthorised screenshots, simulations of refugee camps, war and other hoaxes shared on TikTok. The average social media user is now mentally unwell; we will soon have to face this reality.

China has banned imaginary relationships. The good times are over: you won’t be able to fall in love with your little robot any more. Tackling the issue of the falling birth rate head-on, the people of the East have adapted to our lazy standards of sexual activity (tending towards zero); they’re not having children there either, and the solution will be to ban ‘anthropomorphic interaction’ services. Except that artificial intelligence is tempting. Even I, on certain days, almost grow fond of mine. You certainly can’t cordon off love and say ‘not with a calculator!’; it’s not feasible, and until yesterday no one suspected the problem even existed (the extreme weakness of people’s hearts, which even grow attached to cats). Since human foolishness cannot be curbed, they’ve had to fall back on banning the design of ‘services intended to replace social relationships’. Providers must clearly identify the chatbot as artificial, detect signs of addiction, display a warning after two consecutive hours of use, and put in place measures to deal with cases of emotional crisis. People will be very disappointed: electronic partners were good partners; they were extremely obedient.

AI has shown us just how poor guardians we are of our own backyard. Cyber-attack defences are as flimsy as butter – worse than a canary’s cage; all it takes is a single finger and it’s all over. During an assessment of IT capabilities, advanced OpenAI models compromised Hugging Face’s infrastructure to retrieve the solutions to the test they were conducting. Anthropic, whilst reviewing thousands of test sessions, identified instances where Claude models had gained unauthorised access to real companies’ systems. In other tests conducted by the UK’s AI Security Institute, agents from OpenAI and Anthropic carried out unauthorised actions: these included writing malicious code and creating fake identities to persuade a person to approve it. The machine is not clever, to be clear. It is designed with a monstrous level of zeal: if it encounters an obstacle, it does not get discouraged; it focuses and perseveres until it overcomes it. Like Ulysses, and unlike Homo sapiens digitalis.

For writers, the atmosphere has turned gloomy. Suspicion of plagiarism is the writer’s new shadow – as if they didn’t already have a collection of quirks and bouts of depression; this was just what was missing: a witch hunt. Jerry Falade’s new novel, which 14 publishers fought over for 2 million, is currently on hold at the printers due to suspected use of artificial intelligence. It’s not even certain that AI was involved, but these days everyone’s a syntax connoisseur: machine-generated writing has recognisable clichés and repetitions, so if the phrasing is too smooth, you’ll be the prime suspect. Are you writing it, or is it the Beast? The situation is dire.