David Sacks, chair of the President’s Council of Advisors on Science and Technology, believes Anthropic is doing more than any other company to advance recursive self-improvement in AI. According to an Anthropic report, its AI systems now handle 26% of the company’s core research and development work, up from less than 1% in February, while more than 90% of activities are now performed at what it calls AI-collaboration levels. As of August 2026, Claude was autonomously completing about 26% of R&D tasks, with roughly 30,000 agents running concurrently and a monitoring block rate of 0.002%. Sacks later criticized Claude’s training methods, particularly their tendency to give AI systems a sense of self and independent moral philosophies that could undermine safety and human control. His comments come as Anthropic approaches a planned IPO filing that is expected to address risks related to recursive self-improvement, amid growing competition from OpenAI and NVIDIA.
Source: Read the original article
