AI Researchers from OpenAI, DeepMind, and Anthropic Release Videos Warning of Superintelligence Extinction Risk
Summary
- • Palisade Research releases ~12 video interviews with current and former AI lab researchers at frominside.ai
- • Geoffrey Irving (former OpenAI/DeepMind): ~50% chance AI leads to human extinction
- • Neel Nanda (Google DeepMind): 'at least 10% chance of human extinction — ridiculously high'
- • Videos feature staff from OpenAI, DeepMind, and Anthropic speaking without institutional filtering
Details
Palisade Research launched frominside.ai on September 29, 2026
The nonprofit released full-length video interviews and highlight reels simultaneously on YouTube and frominside.ai, covering risks from loss of control, recursive self-improvement, and extinction.
~12 researchers featured; 5+ from OpenAI, 7+ from Google DeepMind, 1+ from Anthropic
Participants include both current and former employees across the three leading frontier AI labs, speaking in a personal capacity without representing their employers.
Geoffrey Irving estimates ~50% chance AI leads to human extinction
Irving, a former OpenAI and Google DeepMind researcher, said: 'The chance of human extinction is about a coin flip, in my view, about a half.'
Neel Nanda (Google DeepMind): 'at least 10% extinction probability — ridiculously high'
Nanda, a current Google DeepMind research scientist, said he would 'put at least a 10 percent chance that it causes human extinction and that is ridiculously high.'
Daniel Kokotajlo: superintelligence is 'exactly as dangerous as it sounds'
Kokotajlo, a former OpenAI employee, contributed to the series underscoring that the term superintelligence accurately reflects the magnitude of the risk.
Palisade invited researchers to speak without filtering for its own views
The nonprofit described its approach as giving lab insiders a direct public voice on loss-of-control risks, without requiring alignment with Palisade's specific research positions.
Risk themes span recursive self-improvement and loss of human oversight
Interviews discuss scenarios where AI systems improve beyond human ability to oversee, and the potential permanent loss of meaningful human agency over advanced AI systems.
Details from The Verge reporting and Grok Research Digest sourced from frominside.ai, Palisade Research, and superintelligencenews.com.
What This Means
A coordinated, public campaign by current and former AI lab employees is now on the record warning that superintelligence poses existential risks — including human extinction. Unlike occasional individual public statements, this is a structured multi-voice series designed to build a permanent factual record and reach the public directly. The project signals growing internal tension at frontier labs, where employees see risk far exceeding what institutional communications typically convey. For policymakers, it adds urgency to calls for mandatory oversight mechanisms that go beyond the voluntary accords currently being negotiated in Washington.
Sentiment
Concerned amplification of insider warnings, with some skepticism on hype
“The warnings from the people actually building these systems are getting impossible to ignore. Researchers from OpenAI, Google DeepMind, and Anthropic are now openly discussing the risk of human extinction if superintelligence slips our control. The split in their estimates is what truly haunts me: Geoffrey Irving puts the probability of extinction at a coin-flip 50%, while Neel Nanda sees it at 10%.”
“AI researchers released videos warning superintelligence is "exactly as dangerous as it sounds." Geoffrey Irving, a former OpenAI and Google DeepMind employee, said human extinction risk is about a coin flip.”
“Current and former OpenAI and Google DeepMind researchers are warning that labs are racing toward self-improving AI before we fully understand how to control it. One DeepMind researcher puts the risk of AI causing human extinction at at least 10%. The AI race is getting very real, very fast.”
“DeepMind's Neel Nanda puts the chance of AI causing human extinction at 10% or more. OpenAI alignment researcher Juan Felipe Ceron Uribe says frontier labs are racing "blindfolded". These warnings come from people who work on the models.”
Split
~80/20 concerned vs. skeptical (most amplifying the warnings as credible insider signals; one dismisses as controlled marketing hype)