Frontier AI Employees Discuss Extinction Risks and Loss of Control

Summary

Frominside.ai, a project by Palisade Research, presents interviews with current and former employees of OpenAI, Google DeepMind, and Anthropic about their personal views on frontier AI. Several interviewees describe human extinction as a serious possibility, with some offering highly uncertain but substantial personal probability estimates. They discuss possible failure pathways including systems pursuing goals that conflict with human interests, exploiting connected computers, controlling automated factories or robots, using drones, attacking infrastructure, and enabling biological threats. The interviews also address why people who fear these outcomes continue working at AI labs: some believe they can reduce large-scale risks from inside the organizations, while others left over concerns about safety commitments or the need for government oversight. The project says its interviews were unscripted but edited for flow and clarity, and that 22 interviews were filmed, with many awaiting permission for release. It explicitly warns that its participants are not a representative sample: recruitment relied on networks and attracted people who were especially concerned about AI risk or interested in slowing frontier development. The videos are released under a Creative Commons Attribution 4.0 license.