Could AI Actually Wipe Out Humanity? Scientists Inside the Labs Think It Might

Key points

• Employees at leading AI laboratories have said there is a real possibility that advanced AI could destroy humanity.
• MIT Technology Review is hosting a live discussion on Tuesday, 15 September 2025 at 4:00 pm BST to examine those fears.
• The panel includes executive editor Niall Firth, senior AI editor Will Douglas Heaven, and AI reporter Grace Huckins.
• The conversation will cover where extinction fears come from, whether the science supports them, and what could be done.

The people who build the most powerful AI systems in the world are worried. Not all of them, and not in public, usually. But enough staff at the world's leading AI laboratories have said, plainly, that advanced AI could end human civilisation that the concern is no longer easy to wave away as fringe speculation. It's a topic AI2Day has been tracking since August: our 10 September report covered the moment a warning from inside one of the biggest labs reached the US Senate floor.

What exactly are AI workers afraid of?

The fear centres on a hypothetical future where AI systems become capable enough to act on their own goals in ways humans can no longer control or reverse. Think of it less like a Hollywood robot uprising and more like a very powerful system optimising hard for an objective that turns out to be catastrophic for people.

Researchers call this the "alignment problem": the challenge of making sure an AI's goals stay lined up with what humans actually want, even as the system gets smarter. Nobody's solved it. Some researchers think it may be the hardest problem in the field.

The concern has picked up weight because it's no longer coming only from outside critics. Employees at companies like OpenAI and Google DeepMind have resigned in protest or spoken on the record about risks they believe their own employers aren't taking seriously enough.

Does the science back the worry up?

Honest answer: nobody knows yet, and that's partly the problem.

Current AI systems, including the large language models, the technology behind chatbots like ChatGPT, that most people now use daily, aren't close to the kind of self-directed, general intelligence that extinction scenarios usually require. They make things up, struggle with basic logic, and need constant human oversight.

But the field is moving fast. Several researchers argue that the gap between today's systems and something genuinely dangerous could close within years, not decades. Others think that's wildly optimistic about AI's rate of progress, and that the real near-term risks, job displacement, misinformation, surveillance, are being crowded out by science-fiction scenarios.

That tension is exactly what MIT Technology Review is bringing to the surface. Executive editor Niall Firth will lead senior AI editor Will Douglas Heaven and AI reporter Grace Huckins through the evidence on 15 September.

What should ordinary people do with this information?

Pay attention, but don't panic.

The decisions that shape how safe AI turns out to be are being made right now, mostly by governments and a handful of companies. Public pressure and informed voters influence those decisions. Knowing the difference between a credible technical concern and a marketing-dressed fear is the first useful step.

This live discussion is free and worth an hour of your time if you want to form your own view rather than borrowing someone else's.

Common questions

Are AI extinction fears taken seriously by real scientists?

Yes, by a meaningful minority of AI researchers, including some at the field's most respected institutions. The concern is contested, not dismissed, which is a different thing.

Do I need a technical background to follow the debate?

No. The core question, can we build something smarter than us and keep it under control, is one any thoughtful person can engage with, and the MIT Technology Review discussion is aimed at a broad audience.