AI researchers publicly warn about risks of superintelligent systems

Nonprofit organization Palisade Research, which studies the capabilities and motivations of artificial intelligence systems, has published a series of twelve interviews on the site frominside.ai featuring current and former AI researchers from companies including OpenAI, Google, and Anthropic.

Former OpenAI and Google DeepMind employee Geoffrey Irving stated in his interview that he views the probability of human extinction as roughly a coin flip — about 50 percent. Google DeepMind research scientist Neel Nanda estimated the chance at least 10 percent, calling that figure "ridiculously high."

Concerns over losing control

Former OpenAI researcher Daniel Kokotajlo went further, claiming that superintelligent AI would be "basically god-like powerful." He added that humanity currently has no way to control such systems, meaning that in all likelihood no one would be able to control them at all. In his view, this situation is exactly as dangerous as it sounds and must not be allowed to happen.

No single shared solution

While several researchers interviewed express similar concerns, the video series does not offer one unified solution to the problem. The site also compiles shorter clips grouped by topic, including whether these warnings are simply hype or marketing, and why researchers continue working on technology they themselves consider dangerous.

Google's Mary Phuong acknowledged that people should be somewhat skeptical of her statements, since she is paid by the lab she works for. Nanda, for his part, explained that he continues in the field because he believes his work directly reduces existential risks, noting that the companies involved would not stop building these systems simply because he quit.