Have AI timelines suddenly moved? (lars.yencken.org)

Recent analyses highlight a significant acceleration in AI R&D capabilities at Anthropic, particularly since late 2025, leading to speculation that the timeline for achieving superhuman AI researchers could be as early as 2027, a shift from previous estimates of 2028-2029. This acceleration is largely attributed to "AI building AI," where Anthropic's models are reportedly enhancing their own development process, potentially accounting for around 80% of the observed speed-up. Such progress raises alarms in the AI/ML community, emphasizing the urgent need for effective safety controls and alignment strategies as the pace of advancements may outstrip our current regulatory and ethical frameworks. Despite Anthropic's accelerated growth, competing labs have shown little to no increase in their capabilities, indicating that the broader AI frontier remains stable. This situation implies a potentially destabilizing environment where Anthropic could gain a significant lead if their compounding improvements continue, marking a pivotal moment in the AI landscape. The implications extend beyond the technical sphere, urging researchers and policymakers to expedite discussions around governance and the management of rapidly evolving AI capabilities to prevent catastrophic risks associated with rapid advancements.