The Last AI We Control? Inside OpenAI's Dangerous Race for GPT-5

OpenAI's top safety researchers, including co-founder and chief scientist Ilya Sutskever and super alignment co-lead Jan Leike, have resigned amid internal tensions over prioritizing product deployment over safety. Their departures follow a November board crisis where concerns about deployment pace clashed with commercial pressures, leading to the dissolution of the super alignment team tasked with controlling superintelligent AI. These exits highlight a schism between accelerationists pushing rapid scaling for AGI and safety advocates warning of existential risks from unaligned systems.

Researcher Departures and Internal Schism

Sutskever, who shaped OpenAI's technical direction, left in May to found a safe superintelligence venture, forfeiting significant equity. Leike publicly stated that safety processes were deprioritized, with his team denied promised 20% of compute resources amid surging demand for commercial models. This wave of dozens of senior safety staff departures—unlike typical burnout cases—involved specific warnings about governance failures and rerouted resources. They cite a capability curve accelerating faster than safety mechanisms, turning abstract risks into immediate concerns observed in internal capability charts.