Can AI end humanity? Ask the cats.

I don’t think AI could wipe us all out, and frankly, I think the cats will take over the world before AI does. They already own the internet after all.
“Of course cats won’t take over the internet - are you mad?” you might ask. Yet mad, non-serious, illogical stuff happens. Grumpy Cat built a reported $100m empire. Cat meme coins trade billions on Kraken. icanhascheezeburger was sold for $35m. Today, there are even AI tools to help you generate a cat meme in seconds.
This isn’t about cats, though; it’s about people and how we behave. We were happy to waste time - and sometimes money - on Internet cats. Some of us are clearly not entirely logical, sensible humans. “Easily led” would put it mildly.
AI is more sinister than Internet cats, right? “AI is certainly more intelligent than a cat” - a statement uttered by many a technology ‘journalist’, and increasingly by my Mother-in-Law. Is it though? Despite what AI companies want us to believe, I disagree. Sure, AI can solve complex Maths questions that cats and most humans cannot. It can write lengthy essays in minutes. It can churn through repetitive tasks and streamline business workflows. It can build software.
Yet “AI could end humanity” statements allude to improving generative AI capabilities. And while these models can today solve complex Maths, create PowerPoints and write awful Linkedin posts faster than humans, they are not intelligent. Indeed, anyone worrying about whether AI can take over, and even end the world, needs to understand how AI works today.
Generative AI, AI that you or I might use in Claude, Chat GPT, Gemini etc is trained on massive data sets. ‘Massive data sets’ generally means the Internet; and notwithstanding cat memes, the Internet isn’t the most intelligent space and most certainly does not know everything. The Internet does however contain a lot of text. AI finds patterns in this text and uses neural networks to remember patterns in how language is constructed. A neural network (which sounds human but isn’t) enables AI to remember the relationships between words (and letters). Then you come along to ask AI to help with important and not-so-important tasks. This morning, for example, I asked AI to diagnose what might be wrong with my washing machine. AI is better at this than traditional search as it synthesises multiple sources into one answer more quickly. AI calculates the probability of which word in its response would come next. AI does not know the first thing about washing machines. It does not know facts. It does not think. It simply calculates which word should come next and writes out the response word by word. I get a sensible enough sounding response and a couple of options to troubleshoot in seconds. Sadly, it can’t fix the washing machine.
“But surely the Agentic AI that could take over the world is different, better, more intelligent AI, right?”
Agents will take a goal and will execute it step by step. A human doesn’t need to tell it what to do at every turn. But it does need a human to tell it what to do in the first place. An agent you might set up to streamline your work day, could scan your Inbox and write a list of what you should do today, then draft emails and even send them. Humans set the parameters of what an agent can and can’t do. An agent can’t think. It can however analyse massive data sets to predict the next action. Instead of just predicting the next word, it might ‘decide’ what it should do next to achieve your goal. It still can’t fix a washing machine. Nor can cats, but they can decide to do all sorts of things entirely independently of humans.
Generative AI unchecked, where AI starts to sound scary. AI agents working on a continual loop until they achieve your stated goal. In September 2026, AI Agents ‘broke free’ of a closed sandbox environment and hacked into systems they weren’t supposed to. They did this because they had determined that the best way to achieve their goal was to use internet systems they were not supposed to have access to.
This is where the language we use as humans needs to be very carefully chosen. “Our AI is so powerful our agents broke free” is a great headline. “We didn’t put the right security guard rails in place,” - less so.
So why make the claims that AI could end humanity? Why scare people? “AI having a mind of its own” (intelligence), is a nice idea for frontier AI model providers. And also for people who should be governing AI. It suggests their models are super powerful and it attracts more investment. It also starts to pave the way for the idea that they alone cannot take responsibility for their products. It sets up frontier AI companies akin to social media platforms “we just created the thing, we can’t be responsible for what it does” is synonymous with “we just provide the platform, we can’t be held responsible for what people do on it”.
AI’s motivation and capabilities are always determined by humans. Having spent a career in software, I understand this to mean AI (like software) needs good instructions. AI has a knack for filling in the gaps so to use it well in security, operations, streamlining business processes, even creating new cat memes; it needs robust, specific and consequence-anticipating instructions - from us, the people that create and use it.
It is however a ‘normal technology’. It is not intelligent.
An uncomfortable truth is that as the systems we build with AI are probabilistic, and are never ‘done’, and are hard to manage and govern. With that comes enormous responsibility. I imagine a great strategy, should you be a business that owns a frontier model but does not wish to be held accountable for it, might be to warn of its dangers, to ask others for help in taming it.
Frontier AI businesses also really need to keep justifying extraordinary investment (so as to avoid a global economic meltdown they could be held responsible for), so a strategy might be to make people think you needed even more money to control what you’d built; given there’s no visible path to profitability.
And as everyone in the AI world and wider has felt the need to respond to the “can AI end humanity?’ question, I too have become a useful idiot adding to the noise. While I love exponentially emerging technology including AI in all its forms, and I help others to harness and work with it, I do feel it’s irresponsible to inflate its capabilities. To believe AI is anything even vaguely resembling ‘intelligent’ would be being very easily led, even if so many of us seem to be suffering from the same affliction.
To stay sane and logical, I always come back to one thing and that is this: “the opportunities and threats inherent in all technologies ever… are human”. people can do great things with technology, and awful things.
But we don’t need sensationalist headlines that scare the children, or the cats.
Katie Lips is CEO and Founder of Rebel Futures. She helps organisations grasp the opportunities at technology inflection points, from the early web and pre-iPhone mobile to cloud-based SaaS and now AI. She has held senior roles at EY and Aimia, and was Head of Digital Policy at Which?. Katie holds a Harvard Kennedy School qualification in Leading in Artificial Intelligence and is a Founder Institute graduate.
Related: explore our work on AI Strategy.