Neel Nanda, a research scientist at DeepMind, said he believes there is at least a 10% chance that AI could lead to human extinction, a figure he himself described as “ridiculously high.” Coming from a researcher employed at one of the frontier labs actually building the systems in question, the estimate carries different weight than similar figures from academics or advocacy researchers with no direct lab access.
A 10% estimate from inside a major lab is also a useful anchor for a debate that's otherwise dominated by either confident dismissal or confident alarm: Nanda's framing, acknowledging the number is “ridiculously high” while still working inside the field rather than calling for an immediate halt, reflects a middle position that gets less public attention than the more extreme voices on either side.