A scientist at OpenAI wanted a bunker for when AGI arrived. Yes, really.

Concerns over artificial general intelligence (AGI) are no longer confined to science fiction. As major AI labs edge closer to developing AI systems potentially smarter than humans, insiders are voicing increasingly extreme safety warnings. Among the most striking: a proposal to build a doomsday bunker in preparation for AGI’s release.

Scientists at OpenAI feared an AGI-triggered catastrophe

In 2023, during a high-level meeting at OpenAI, then-chief scientist Ilya Sutskever suggested constructing a bunkerto shield researchers from the fallout of an AGI event. The idea, cited in Karen Hao’s upcoming book Empire of AI, was reportedly discussed multiple times within the company, reflecting deep-rooted fears among OpenAI staff about AGI’s impact on humanity.

Sutskever, now founder of Safe Superintelligence Inc., was a central figure in developing some of OpenAI’s most advanced models, including ChatGPT. His remarks weren’t made in jest: colleagues confirmed his repeated insistence on building a secure facility, treating the arrival of AGI as a potential existential risk.

Warnings grow louder as AGI approaches

This unease is shared by other industry leaders. Google DeepMind’s CEO Demis Hassabis recently admitted that AGI developments are keeping him awake at night, while Anthropic’s CEO Dario Amodei confessed that his team doesn’t fully understand how its models work. Such statements underline a disturbing truth: those closest to AGI’s creation are among the most alarmed by its power.

Despite OpenAI CEO Sam Altman’s belief that AGI will arrive without immediate disruption, internal anxiety reveals a starkly different perspective, where bunkers and existential dread are part of daily conversations. Whether hyperbolic or prophetic, Sutskever’s bunker idea may prove a chilling symbol of our uncertain future.