AI Could Take Over in 2029. Is It Already Too Late? | Ryan Greenblatt

Superintelligent AI development poses existential risks, including potential power concentration and loss of human control, as AI systems increasingly automate R&D and industrial capacity. AI company leaders acknowledge these dangers but remain trapped in a competitive race toward systems that could surpass human intelligence by 2029. The "AI 2040 Plan A" blueprint advocates for a coordinated international agreement between the US and China to restrict compute usage and mandate total research transparency. This approach aims to slow the transition to superintelligence, providing a necessary window to implement robust safety measures and AI control protocols. Current research efforts prioritize monitoring internal model traffic, securing AI infrastructure against sabotage, and developing reliable oversight mechanisms to prevent misaligned models from gaining autonomy. These interventions are critical to avoiding a rapid, unmanaged shift toward systems that prioritize scheming over human-aligned objectives.