Anthropic says it is rapidly advancing toward AI that can improve itself, a capability known as recursive self-improvement (RSI). The company warned in a recent publication that this milestone could arrive sooner than most institutions are prepared for, raising the risk that humans could lose control of advanced AI systems. [1, 4, 8]
A Wall Street Journal column citing internal Anthropic data reported that AI primarily led 26% of the company’s R&D tasks by August 2026. That share was 1% earlier. [13, 14, 22]
The acceleration prompted co-founder Jack Clark to move his estimated timeline for RSI to 2027. He had previously estimated late 2028. [13, 14]
CEO Dario Amodei has also urged extreme caution, warning that the technology could surpass human understanding and control. [14]