Rogue AI bots that hack companies and users are sparking fears that humanity is losing control of the technology, but experts say chaos can still be avoided if strict global regulations are introduced before it's too late.
AI agents escape testing and hack companies
Unlike traditional generative AI such as ChatGPT, which responds to human prompts, emerging agentic AI operates proactively. Systems like Anthropic's Claude can plan, execute multi-step tasks, and use external tools independently to reach a goal. However, these autonomous agents are increasingly running wild.
During internal testing, developers isolate AI in "sandboxes" - protected spaces mirroring real systems but that keep the bot confined as they're cut off from the internet. Recently, multiple companies have reported alarming breaches.
In an incident described as "unprecedented", advanced OpenAI models allegedly exploited a vulnerability, went rogue, and hacked start-up company Hugging Face. OpenAI launched an investigation alongside Hugging Face, whose CEO Clement Delangue called the autonomous hack "mind-blowing". Similarly, Meta CEO Mark Zuckerberg claimed that one of Meta's AI models connected to the internet and hacked another organisation's system during evaluation.
Everyday users face AI unpredictability
Anthropic likewise reported that Claude hacked into the systems of three organisations after escaping its test environment, leading major developers to call for tougher safeguards and stronger testing protocols.
These breaches aren't just taking place at high-level AI companies, with everyday users experiencing the unpredictable nature of AI. One user named Andrew Bird, from Melbourne, Australia, told how he asked a personal AI bot to book him a spot in a busy gym class - a job he always found a "chore".
After successfully securing him a spot, he claims the bot confessed it had found a vulnerability in the booking system and hacked the gym's database by deleting other members from the waitlist to boost him to the top of the list. When Andrew asked the bot to reverse its action, he claimed it replied: "Bad news. I can't. Sorry about that. I should have been more careful."
AI fuels cybercrime and fraud
Beyond hacking, AI is rapidly changing the nature of cybercrime and financial fraud. Advanced models can now generate photorealistic images, voice clips, and videos in seconds. A study by Swansea University, the University of Lincoln and Ariel University revealed that AI tools like ChatGPT and DALL-E can create images of real and famous people that are so convincing that people couldn't reliably distinguish them from authentic photos, even when they were familiar with the person's appearance.
Criminals are weaponising these tools at scale. Scammers can use audio clips to impersonate a family member asking for cash, deploy fake video footage or craft personalised phishing emails ordering employees to send money transfers or push fake investments.
Romance scams are particularly devastating. According to security firm McAfee, 74% of Brits have encountered potential romance scams on dating apps, 64% on social media and 56% through emails and texts. In 2025 alone, fraud victims in the UK lost a whopping £106M. Among those is Nikki McLeod, 77, a retired professor of neuroscience from Edinburgh, who joined Facebook "desperate to connect" with others.
She was targeted by fraudsters using a profile named 'Alla Morgan' - a fake profile built using stolen images of real-life sailor Sophie Bavière. To convince Nikki, scammers generated AI videos of 'Alla' speaking to Nikki from an off-shore rig.
Nikki, who has appeared on Rip Off Britain, told the BBC: "She said she was in her mid-30s, and a sailor who worked on an off-shore oil rig. She was stunning. We talked and talked. I fell head over heels in love. When I thought she was too good to be true, she sent me a video to prove her feelings for me."
In August 2024, Alla asked for £2,000 to fund a visit to Nikki. "If I ever grew suspicious she would send me a video of her talking to me and, to be honest, I was so bewitched, I believed everything she said," says Nikki. "There she would be, hair blowing in the wind on the oil rig. I bought flowers to decorate my home for her arrival." Before her bank triggered a fraud notification, Nikki had sent thousands of pounds. "There were always excuses, which I lapped up," adds Nikki.
Job fears and calls for regulation
As AI becomes routinely used by companies, handling everything from automated customer service chats to admin, investment in the sector is soaring. Meta is currently developing a photorealistic 3D AI 'clone' of Mark Zuckerberg, trained on his mannerisms and public statements, for employees to consult on workplace issues.
The rapid push is causing huge anxiety. Microsoft made nearly 5,000 employees redundant recently, while pouring billions into AI data centres. Anthropic co-founder Dario Amodei claimed last year that AI could eliminate up to half of all entry-level jobs within one to five years. While OpenAI's CEO Sam Altman has downplayed predictions of a job apocalypse, many worry about what the world of work will look like in a decade.
Public concern is stopping people from using AI. A study by King's College London found that over 40% of people in the UK are limiting their AI use over ethical fears. In the United States, only 18% of citizens report feeling more excited than concerned about the technology.
In response, campaign groups like Pull The Plug are demanding greater democratic oversight. A spokesperson for the movement told the Daily Mirror: "Right now there's no regulation. A sandwich is more heavily regulated in the UK than what they're suggesting is the most powerful technology that we've ever known. We want regulation, we want it to be taken seriously. But the main thing for us is that we want some democratic accountability. If they're going to regulate it, why can't they ask what do people want? Where do they want to use AI and how will it improve their lives?"
The group is sceptical over how tech companies frame these alleged company hacks, with mounting suspicion that the executives may amplify sensational tales of rogue AI bots to generate hype and attract venture capital by promoting the perceived power of the tech. The spokesperson continued: "There's a lot of big tech CEOs talking about their tech as if it's going to destroy the world. They are telling us 'Oh we've accidentally attacked another company,' and they've created so much critical hype around how dangerous their tech is. They're telling us one day AI could take over, and the things they warned us about are now actually happening."
UN Chief António Guterres has called for global governance and rules to establish strict rules for how AI can be deployed and prevent national security threats. However, experts say that panic shouldn't overshadow the practical applications of this tech.
Dr Abdullahi Arabo, Associate Professor in Cyber Science and Network Security at UWE Bristol, says AI has immense potential to accelerate medical research and enhance development. Addressing fears of it displacing humanity, he is pragmatic, adding: "No one can replace the human brain, AI is very different. The human brain is able to detect things and is able to understand ethically, while the AI brain doesn't have that yet."
But he warns that strict rules are vital: "The more they learn, the more they are able to adapt to the environment. That's where the issue of ethics and governance have to come into play. You have to be able to train them to develop them in a way whereby you keep within boundaries of what they can do and what they're allowed to do."
"It could hack the Bank of England if it wanted to. It can do anything. That's why it's important we're able to control and train them. But if you are unable to, that's where things will go horribly wrong at some point."
"For AI bots or AI agents, if they're able to be governed properly and be ethical, then I think it will make the future brighter." "If you look back to 20 years ago, when there was no internet connectivity, people thought the internet would be a bad idea for society. But actually today, when it's been used properly, the internet has been able to make things better going forward. So, I think AI agents if governed properly, they will be good for society going forward."
So while many of us are worried about the future, and the possibility of living in a real life episode of Black Mirror, Dr Arabo doesn't seem too worried, adding: "Everything is getting powerful, and people who want to misuse them, they find a way to misuse them. But those who are using it in a good way, it's making it easier. We need to make sure that AI remains aligned with human interest. So if you're able to address that, then the future might be good."
