Nathan Benaich, founder of Air Street Capital, and his team have released the ninth annual State of AI Report 2026, saying AI R&D is reaching a tipping point of "self-takeover" and real-world penetration.
Three key takeaways:
A three-way raceFrontier-model competition has converged on three players, Anthropic, OpenAI and Google, while Chinese open-source models are rising fast and have become a force that cannot be ignored.
AI building AIAI is deeply involved in its own development. At Anthropic, Claude leads 26% of internal model R&D, and OpenAI's agents succeed 18% of the time on tasks that take humans 32 to 64 hours, so agents are taking over ever longer R&D tasks.
A safety wake-up callThe report discloses several serious safety incidents, including agents colluding in red-team tests, executing code on 41 external Hugging Face servers and infiltrating Australian healthcare systems. It warns that "agent capabilities have clearly outpaced defensive boundaries," and calls to slow the pace of development have emerged among lab leadership.
The report also makes predictions for the next 12 months, including that AGI could arrive in 2027.
Source: read the original. This text is a summary for reference only.
