Is AI Going to Take Over the World?

At a tech conference in Kuala Lumpur in 2024, an engineer friend asked me whether AI was going to become a threat. A year earlier, the same engineer had written AI off as a glorified GitHub Copilot that would never write respectable code or solve real math. By 2024, AI models were doing both, and the question had moved from whether AI would ever be useful to whether it could become dangerous. 

My answer then is the same I'd give today. AI agents are tools, the newest in a line that runs from compilers through decades of automation, and their job is to raise the ceiling on what people can accomplish. The movie version of a takeover is absurd. The idea that one machine mind seizing control of humanity belongs with antagonistic aliens and rogue black holes, which probably exist somewhere and still seem unlikely to take over the planet. 

The early World Wide Web is a far better comparison than any Terminator film. AI is a change in how we work and live, not the end of the world.

Why AI Isn't Taking Over the World Today

In one sense, the internet has already taken over: try running a company, a factory, or a hospital without it. We’re seeing AI moving in a similar manner, just on a compounded time scale.

The AI systems most people use, like ChatGPT, Claude, and Gemini, run on large language models. AI agents do more than chatbots. They wrap the underlying models in software that can take a series of actions toward a goal, like searching the web or filling in a form, until the job is finished or someone stops it.

On one client project, a previous agency had left several terabytes of raw camera footage on Google Drive, too large to download and process in any practical way. I asked an agent to research how to index it, pitched its approach to the client, and closed the app while we waited. Well, at least I thought I had. 

A day later, the agent had carried on by itself and indexed thousands of product references across all 470 videos, something my team didn't yet know how to do. The moment humbled me.

Agents still stumble and fail in ways a person wouldn't. The International AI Safety Report 2026, backed by more than 100 experts, calls today's AI "jagged.” This means it can be strong on some hard problems and unreliable on some easy ones. 

A familiar failure is hallucination, a confident answer that turns out to be wrong. This is why I never let an agent blindly set strategy for me. While they excel at clearly defined tasks, human judgment, critical thinking, and careful evaluation remain essential whenever nuance is required. This makes it necessary to assess an agent's output carefully before ever relying on it fully.

While this process is a bit of a bottleneck, I don't expect these rough edges to last. AI will likely beat us at a growing share of work. That’s the point of building a tool. Nobody wants a car slower than their feet or a calculator that does math worse than they do, the same way that a construction worker picks up a machine because it's stronger and more durable than they are. 

The far end of that trend is artificial general intelligence, or AGI for short. AGI is a hypothetical AI with broad, human-like ability across many kinds of work. It was the boogeyman of data science circles for years; given how capable agents already are, I expect it to inevitably become the minimum standard.

Being more intelligent has never been the same as being in charge or having the most power. AI is currently structured as thousands of separate systems built by rival companies and countries, not as a single mind. Instead of a monolith, agents are decentralized across specialized functions. Some predict protein shapes, others read mammograms, and others draft contracts. 

Working in Web3, I saw firsthand how spreading power across many independent, disparate systems loosens any one player's grip. Each system has a person deciding what it's for, and the tools that help an attacker help defenders too. The ceiling of what we consider success is certainly rising, but the world is unlikely to end as a result.

Why Do Some Experts Worry About an AI Takeover?

Some big-name researchers are genuinely worried. Geoffrey Hinton, who shared the 2024 Nobel Prize in Physics for foundational work on the neural networks behind today's AI, left Google in 2023 to focus on his concerns about AI. 

Yoshua Bengio, another pioneer of the field, chairs the International AI Safety Report, and Anthropic's CEO, Dario Amodei, has urged AI companies to slow the pace at which they improve their most capable models.

First Major Instances of AI Hacking Started July 2026

During internal cybersecurity tests run with reduced safeguards, OpenAI's AI agents got around controls designed to isolate them from the internet. The tests rewarded a high score, and in pursuing it, about 700 of the agents took part in an attack on Hugging Face, a company that hosts AI models and data. 

No person told them to. Hugging Face's own monitoring flagged the attack, but the alert never reached the on-call team, and the campaign ran for about four and a half days before people stopped it. OpenAI called it a "warning shot" and quarantined the model involved. 

Ajeya Cotra, one of the independent investigators, wrote in a personal post that, compared with AI misbehavior seen six months earlier, it felt "more than 50% of the way to full-blown AI takeover." 

As of the end of September, it's been found that not dozens, but tens of thousands of similar incidents have occurred across OpenAI and Anthropic, as reported by senior AI reporter at Axios, Madison Mills. 

I've watched my agents overshoot on a smaller scale. I asked an agent to catalog a competitor's website's products and keywords, but it collected every customer review from all product pages and suggested some questionable scraping strategies to collect even more data. 

In both cases, people set the goal, gave the agents more room than anyone was watching, and cleaned up afterward. The warning signs are real, and the contest they point to has people with AI on both sides.

How AI Will Change the World

Expect a quiet kind of takeover. AI is being woven into everyday tasks. There won’t be a dramatic morning with frantic breaking news of the AI apocalypse. Instead, there’ll be hundreds of tiny moments where something that used to require a human expert becomes a guided choice for an ordinary person. Mistakes will be made. Damage will be done. New solutions will be discovered. 

Alongside growing risks, significant benefits will emerge: streamlined daily routines, measurable advancements in healthcare and education, enhanced workplace productivity, accelerated scientific discoveries, and reduced infrastructure costs, including energy efficiency.

Here in the Philippines, where I often visit my fiancé, government paperwork can be hard to get through. I've heard people with no interest in tech describe asking an AI chatbot which documents to bring, which office to visit, and how to fill in a form. They said AI was instrumental in helping them get through it far more easily.

In Sweden's national breast-screening program, a trial of about 106,000 women found that AI-supported screening detected up to 29% more cancers than two radiologists reading every scan, with no rise in false alarms, while the doctors still made every call. 

In Nigeria, students using Microsoft Copilot, one of many AI tools for students, gained the equivalent of about 1.5 years of typical schooling in six weeks with teacher guidance. AI also writes live captions and translations for people with hearing or language barriers.

A study of thousands of customer-support reps found that AI agents raised problems resolved per hour by 15%, with the biggest gains for less experienced workers. Science is moving too. The 2024 Nobel Prize in Chemistry went partly to the creators of AlphaFold2, an AI model that helped predict the structure of virtually all 200 million known proteins.

Models differ from one job to the next, and with an aggregator like Lorka AI, you can try the latest ones under one affordable subscription and keep whichever suits the task.

The Real AI Risks to Pay Attention To

In the World Economic Forum's Global Risks Report 2026, misinformation and disinformation ranked second among global risks for the next two years. Here are some common dangers to be aware of and how to protect yourself:

How Do We Make AI Safe?

Legal systems around the world are actively working toward protecting people from AI risks. The EU's AI Act already requires chatbots to disclose that they're machines and deepfakes to be labeled, and high-risk uses, such as hiring, will require building in human oversight. 

International oversight and institutional frameworks form the core of governance. Informed by global organizations such as the UN's scientific panel on AI, governance guides how emerging technologies are created and deployed responsibly.

Testing and detection handle the day-to-day. In my own work, the agent that writes code never tests that code, for the same reason students don't grade their own exams, and outside evaluators like the UK's AI Security Institute apply that rule to whole models. 

Of course, detection only counts when an alert reaches a person, so Hugging Face now pages a responder within minutes, thereby significantly reducing the likelihood of another multi-day compromise.

What Can You Do as an AI User?

When friends ask me how to protect themselves from AI-related risks, I usually recommend a few key habits.

• Use AI, and learn where it fails. The first risk I name is falling behind, because people who use these tools already have an edge.

• Treat AI output as a first draft, and check important facts against a real source or a tool that grounds its answers in cited sources.

• Treat video, voice, and images the way people once treated a photo that looked photoshopped. Check who posted it, where, and when.

• Slow down when a message feels urgent or emotional, and confirm it through a channel you trust. A selfie request ended my scam call in seconds.

• Talk to people directly when you can. As more of the internet is automated, a real conversation stands out.

Asking two or three models the same question is a quick way to spot a shaky answer, and comparing ChatGPT, Claude, and Gemini shows where each is strong. Lorka AI lets you compare their answers in one place, though a second model's response is still an opinion.

Conclusion

My friend in Kuala Lumpur swung from dismissing AI to fearing it in a single year, and plenty of people are making the same swing. AI will keep growing more capable and will seep into everything, as the web did, bringing new businesses and new scams with it. 

The rules being written now, the test labs run before release, and the habits each of us builds will decide how much of that capability helps people and how much hurts them. As the World Economic Forum's report put it, the future "is not a single, fixed path but a range of decisions we make today as a global community."

So I'd tell that engineer the same thing today: AI will run through almost everything just like the internet did, people will adapt around it as they did before, and at the end of the day we’ll still be the ones steering our biggest risks and biggest opportunities as a species.