Anthropic filing shows AI's share of its R&D work at 26% in August

Anthropic's listing filing included a chart showing that AI performed 26% of the company's research and development work in August 2026, up from about 1% in March 2026, according to digitaltoday.co.kr. The same filing states that the share of R&D work handled by AI without human intervention stood at 0% as of August 2026.

The chart was itself produced by Anthropic's Claude model. Claude agents reviewed 15,000 internal AI R&D tasks and graded them on a six-step scale, from no AI use to fully autonomous operation.

The figures have revived discussion of recursive self-improvement, or RSI, in which AI systems take part in building their successors. In the filing, Anthropic CEO Dario Amodei said uncontrolled RSI could outpace humans' ability to understand and control it, and that it may need to be avoided altogether.

Other labs set their own targets

In September 2026, OpenAI announced research intern automation capable of carrying out tasks that take skilled researchers days, and set a goal of an automated AI researcher by March 2028. Anthropic co-founder Jack Clark now expects RSI around 2027, moving up from an earlier estimate of the end of 2028.

At xAI, Elon Musk said in March 2026 that human involvement in improving the Grok model keeps shrinking and that new models are all made by previous models. In a March 2026 report by the Wall Street Journal, OpenAI's chief scientist called RSI the only way forward, while an Anthropic co-founder called it possibly the most important technology trend.

Not everyone welcomes the direction. In a September report by Fortune, Anthony Aguirre, head of the Future of Life Institute, called AI increasingly taking over research "perhaps the worst idea in human history."