"Model progress is now extremely rapid, and we always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment."

That’s exactly how I feel this year, too. 2026 feels like the year in which the pace of model development truly accelerated.

Models are improving remarkably quickly. Open-weight models are becoming competitive with proprietary frontier systems on some benchmarks, while a growing number of developers are releasing increasingly capable models in rapid succession.

Amodei and Anthropic have repeatedly pointed to late 2026 or early 2027 as a possible arrival window for what they call “powerful AI” (superintelligence). Considering the progress we’ve seen this year, alongside the first cases of frontier labs delaying or restricting access to models because of safety concerns, that timeline feels increasingly plausible.