Palisade Research has published interviews with current and former employees of leading AI labs at frominside.ai. The series, reported by Reuters on September 29, asks whether humans can retain control of much more capable AI.
On the original site, DeepMind researcher Mary Phuong warns that rapidly growing capabilities, combined with inadequate control over model motivations, could lead to loss of control. Her colleague Neel Nanda gives a personal estimate of at least ten percent for the risk of human extinction from AI. That is not a measured probability or company forecast.
OpenAI researcher Juan Felipe Cerón Uribe and former OpenAI and DeepMind researcher Geoffrey Irving also participate. The project links longer interviews and attributes short excerpts to individual speakers and timestamps. Participants speak personally, not for their employers.
Palisade explicitly describes its selection as unrepresentative: its own networks and particularly concerned safety researchers are overrepresented. Interviews were edited for flow and clarity, sometimes combining takes. This risk signal documents the publication and its warnings, not a new AI incident or an industry-wide consensus.