Anthropic leads three-lab AI frontier: Claude does 26% of its R&D

The report is framed as an annual stocktake rather than a single product announcement. Its author traces his interest in AI to a PhD in cancer research and computational biology in the early 2010s, and to investing since 2013 in companies building and applying AI — the perspective behind his argument that the gap between people getting substantial value from AI and those getting little is a "human skill issue." That framing sits behind his call for a Genius Bar for AI, where people can get hands-on help choosing and setting up tools.

The report also grades its own past calls, which gives its current claims a track record to weigh. It records two hits, five partial outcomes and three misses, and notes earlier predictions that NVIDIA would fail to acquire Arm, that transformers would achieve leading results beyond language, and that an open-source model would surpass OpenAI's o1 — which DeepSeek-R1 subsequently did. The three predictions offered now follow that same format, including an autonomous AI team beating human-led model research on equal time and compute, and US AI labs officially launching frontier cyberdefense products.