Editorial illustration for AI Researchers Warn Superintelligence Is ‘Exactly as Dangerous as It Sounds’
AI Researchers Put 50-50 Odds on Human Extinction
AI Researchers Warn Superintelligence Is ‘Exactly as Dangerous as It Sounds’
Geoffrey Irving spent years working inside OpenAI and Google DeepMind. Now he's putting a number on the odds that AI wipes out humanity: roughly 50-50. He's not alone in saying it out loud. A new project called frominside.ai, run by the nonprofit Palisade Research, has collected a dozen video interviews with current and former employees of OpenAI, Google, and Anthropic, and released them under a title that doesn't leave much room for interpretation: superintelligence is "exactly as dangerous as it sounds."
The people talking aren't outside critics or academics with no stake in the technology. They built it, or still do. Neel Nanda, a research scientist at Google DeepMind, is among them, putting his own extinction estimate at 10 percent and calling even that "ridiculously high." Former OpenAI researcher Daniel Kokotajlo shows up too, with warnings about what happens once systems outpace human control.
None of them agree on what to do about it. That's the harder problem sitting underneath the headline numbers, and it's one the "AI safety" label has never quite managed to solve.
Former OpenAI researcher Daniel Kokotajlo went as far as claiming that superintelligent AI “would basically be god-like powerful,” and added, “unfortunately we don’t know how to control them at all. And so probably nobody would control them. This is exactly as dangerous as it sounds and must not be allowed to happen.”
Why this matters
We'd normally discount coin-flip extinction odds as clickbait, except these aren't fringe figures. Geoffrey Irving worked at OpenAI and DeepMind. The dozen interviewed on frominside.ai include current employees at the same labs racing to ship the technology they're warning about.
That contradiction is the story: people cashing paychecks from OpenAI, Google, and Anthropic are also telling the public their employers might be building something that kills everyone. For developers and founders building on top of these models, that's worth sitting with, not dismissing as PR theater from a nonprofit called Palisade Research. Whether their capability assessments hold up is a separate question from whether they believe what they're saying, and the interview format, on-camera, named, on the record, makes casual dismissal harder than the usual anonymous "researchers say" hedge.
Watch whether any of these researchers follow words with resignations, product delays, or internal petitions at their own companies. Statements are cheap. What happens next inside OpenAI, DeepMind, and Anthropic is the actual signal.
Related Reading
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
From Archives NVIDIA's AI Development Constraint: Isolation as a Scaling Unit
From Archives BBC's Re-Released 2000 Exhibition Shows How We Imagined Internet's Future
Common Questions Answered
What probability does Geoffrey Irving assign to AI wiping out humanity?
Geoffrey Irving, a former researcher at OpenAI and Google DeepMind, estimates roughly 50-50 odds that AI could wipe out humanity. His willingness to publicly state these extinction-level risks reflects a growing concern among AI safety researchers about superintelligence dangers.
What is the frominside.ai project and who runs it?
frominside.ai is a project run by the nonprofit Palisade Research that has collected video interviews with current and former employees from OpenAI, Google, and Anthropic. The project documents insider perspectives on superintelligence risks, with the release titled to emphasize that superintelligence is 'exactly as dangerous as it sounds.'
According to Daniel Kokotajlo, why is superintelligent AI considered dangerous?
Former OpenAI researcher Daniel Kokotajlo claims that superintelligent AI would be 'god-like powerful' and that researchers currently lack the knowledge to control such systems. He emphasizes that this uncontrollability makes superintelligence exactly as dangerous as it sounds and must be prevented from happening.
What contradiction does the article highlight about AI safety researchers?
The article points out that current and former employees of OpenAI, Google, and Anthropic are simultaneously collecting paychecks from companies racing to develop AI technology while publicly warning that their employers might be building something capable of killing everyone. This contradiction between employment and public warnings about existential risk is presented as the core story.
Further Reading
- AI researchers put out videos saying superintelligence is ‘exactly as dangerous as it sounds’ - Headtopics
- AI researchers put out videos saying superintelligence is ‘exactly as dangerous as it sounds’ - Ground News
- Who's Who In The Fight Over Whether AI Will Kill Us - Forbes
- We Asked Experts If an AI-Driven Apocalypse Could Happen and How to Stop It - Business Insider
- AI researcher says there's a way to avoid existential threat to humanity - Yahoo News
