Sakana AI has launched a dedicated research lab focused on recursive self-improvement, according to The Decoder. The Japanese startup, co-founded by Transformer co-author Llion Jones, is exploring systems that can iteratively improve themselves rather than relying only on ever-larger compute budgets.
The idea could become an important alternative path in frontier AI research, but it also raises control and safety questions. If AI systems become better at improving their own capabilities, researchers and regulators will need stronger methods for evaluation, containment, and governance before such approaches are deployed broadly.