Core Event: AI Safety Researchers Release Collective Warning Videos
Palisade Research, a nonprofit studying AI capabilities and motivations, publicly launched the ‘frominside.ai’ platform featuring in-depth interview videos with current and former researchers from OpenAI, Google DeepMind, and Anthropic. The videos center on existential risks posed by superintelligent AI, with participants explicitly stating that human extinction is plausible. All videos are freely accessible without subscription or paywall.
Key facts:
- Publisher: Palisade Research (nonprofit)
- Platform: frominside.ai (public access only)
- Participants: 12 researchers from major AI labs
- Former OpenAI researcher Geoffrey Irving, Daniel Kokotajlo
- Google DeepMind research scientists Neel Nanda, Mary Phuong
- Format: Full-length interviews plus thematic short clips
Risk Assessment: From 10% to 50% Extinction Probability
Multiple researchers presented extinction probabilities far exceeding public perception. Google DeepMind’s Neel Nanda stated: “At least a 10 percent chance that AI causes human extinction, and that is ridiculously high.” Former OpenAI and Google DeepMind employee Geoffrey Irving offered an even starker assessment: “The chance of human extinction is about a coin flip, in my view"—implying approximately 50% probability.
Most alarmingly, former OpenAI researcher Daniel Kokotajlo described superintelligent AI as “basically god-like powerful” and added, “unfortunately we don’t know how to control them at all… This is exactly as dangerous as it sounds and must not be allowed to happen.”
Unexpected contradiction: Despite these grim projections, nearly all interviewees remain employed in the very field they warn about. Nanda explained: “If I did not believe that my work was directly reducing these existential risks, I would quit… These companies are not going to stop making these systems just because I quit.” This highlights a core tension—researchers acknowledge fatal risk while lacking leverage to halt progress.
Ethical Dilemmas: Trust Deficit and Motivation
The interviews also address longstanding credibility challenges in AI safety research. Google’s Mary Phuong voluntarily admitted: “I think you absolutely should be suspicious of what I’m saying because I am being paid by the lab.” This self-awareness signals recognition of inherent bias.
Scholars further noted the conceptual ambiguity defining “AI safety” itself. As reported, the community lacks consensus on what the term encompasses or what concrete measures qualify as safety work, fragmentation that impedes coordinated action.
Video clips are organized by theme on the site, covering:
- Whether risks are exaggerated
- Personal motivation for working on high-danger technology
- Tensions between individual ethics and organizational goals
Recommendations for Readers
For AI practitioners:
- Systematically study potential existential dimensions of your work, especially interpretability and alignment challenges
- Evaluate whether your organization’s safety governance includes transparent, independent audit mechanisms
For general audiences:
- Current frontier models from leading vendors have not reached the “uncontrollable superintelligence” stage described in these warnings; present risks remain focused on bias, misuse, and error
- Access original videos on frominside.ai for unfiltered perspectives rather than relying on secondary summaries
Final Thoughts
When leading AI safety researchers are themselves issuing worst-case warnings, it represents a watershed moment. When those closest to the technology are also its most vocal doomsayers, the discourse has shifted from speculation to internal reckoning. What’s needed, as these candid exchanges suggest, is not technological retreat—but verifiable, falsifiable safety commitments embedded in research practice.
Palisade Research’s release marks the moment where AI safety concerns transition from academic fringe to insider consensus: the risk is real, the control problem is unsolved, and urgency is justified.