Self-Improving AI Is the Key Step Toward AGI, Says Z.ai Founder — With Safety as the Prerequisite

The founder of Z.ai has publicly argued that self-improving AI systems represent the critical technical threshold on the path to artificial general intelligence, while emphasizing that safety infrastructure must precede any deployment of such systems. The statement positions self-improvement — where models iteratively refine their own weights, architectures, or reasoning processes — as the defining capability gap between current frontier models and AGI. For researchers and developers tracking the AGI timeline debate, this framing from a lab founder adds a practitioner perspective to what has largely been a theoretical discussion. It also signals that Z.ai is actively working in this direction and views safety not as a constraint but as a prerequisite for viable self-improving systems. Developers building on top of frontier models should pay attention as self-improvement capabilities begin to influence how labs design API access, rate limits, and safety guardrails.