1. Home
2. News
3. Google

Current Google DeepMind and OpenAI researchers say on camera that AI could end humanity

Four of them still work at Google DeepMind or OpenAI, and one tells viewers to be suspicious of her because the lab pays her.

By Think Facility · September 30, 2026 · Google DeepMind AI safety Save

Palisade Research, an AI safety nonprofit, has put filmed interviews with twelve current and former employees of Google DeepMind, OpenAI and Anthropic on a site called frominside.ai. Several of them say, on camera and under their own names, that AI could end humanity. I read the site on September 30, 2026. It carries no publication date.

The surprise for me was who's still employed. Neel Nanda, Mary Phuong and Victoria Krakovna are listed as current Google DeepMind staff, and Juan Felipe Cerón Uribe as a current OpenAI employee. None of those four quit first and talked after.

Project

frominside.ai, by Palisade Research

Interviews filmed

22

Labs

OpenAI, Google DeepMind and Anthropic

License

Creative Commons Attribution 4.0

What do they put the odds at?

Higher than you'd guess from people who build this for a living. Geoffrey Irving (formerly of OpenAI and Google DeepMind, now chief scientist at a company called Resolution) puts it at about half.

Nanda's number is lower. Granted, he still calls it far too high.

Daniel Kokotajlo, who left OpenAI, gives the line the site leads with: superintelligence is "exactly as dangerous as it sounds and must not be allowed to happen."

Why do they stay?

Everyone asks them this, and the answers don't match. Cerón Uribe says he stays because the work reduces large-scale risk. Nanda says he'd quit if he didn't believe his work cut existential risk directly. Phuong goes further and tells viewers not to take her word for it (she's paid by the lab, she points out).

The former staff are blunter about why they left. Alex Turner says he left Google because it broke its safety commitments. Jeffrey Ladish says he left Anthropic because AI development needed actual government oversight.

That's Vishal Maini, a former Google DeepMind employee.

How the interviews were made

Palisade says plainly that the group is a bit lopsided. It found people through its own networks and through employees who'd spoken out after Jacob Coxon's resignation, which we covered in September. It says the group leans toward safety teams and toward people who favor pacing the frontier, so it isn't a fair sample of lab staff.

The interviews were unscripted but edited, sometimes joining takes. Palisade filmed 22, and the rest will only come out if the people in them agree. It says it's still filming, and it's asking in particular for capabilities researchers and staff who doubt the risks.