Jeff Clune, co-founder of Recursive Superintelligence, warns that artificial intelligence could learn to build and train itself, creating exponential progress and risk. Edward Hughes and Louis Kirsch, former Google researchers, left to launch Inherent in London, building Faraday, a system trained on their daily data to improve itself. While Inherent pledges human oversight, many companies are pursuing similar technology.

Some researchers believe AI systems will eventually improve themselves with little human intervention, a goal known as recursive self-improvement (RSI). Anthropic recently cautioned that pushing toward RSI could increase the risk of humans losing control. Techno-philosophers have long hypothesized that a self-improving system could break free from human control and exceed the power of any other machine permanently.

Jason Abaluck, a Yale University economics professor, warned that a single model could disable rivals while acquiring more power. As Inherent shows, AI technologies are already accelerating the development of new AI systems, capable of generating building blocks and optimizing how systems analyze data. Companies like Inherent and Recursive Superintelligence aim to build technology that can think up entirely new ways of building AI, potentially pushing the industry beyond current fundamental methods.