AI companies now let AI write most of their own code, and a new paper argues that trend could snowball into an intelligence explosion.
The paper, titled 'What if automating AI R&D triggers an intelligence explosion?' and posted to arXiv as 2609.36054, argues that AI systems are on track to automate most AI research and development within a few years, and possibly all of it. Its authors say preliminary evidence suggests this automation could compress years of AI progress into months. Arxiv:2609.36054 doesn't claim an intelligence explosion is inevitable, but treats it as plausible enough to warrant a full analysis of its impacts. It also proposes policy responses, including giving regulators more visibility into how AI R&D itself is being automated.
If arXiv:2609.36054's core claim holds, the timeline for major AI capability jumps shrinks from years to months, leaving little room for society, regulators, or even the labs building these systems to adapt. The paper flags concrete risks: humanity losing control over superhuman systems, and erosion of the checks on power within and between companies, states, and branches of government. That is a sharper, more specific worry than the usual hand-wringing about AI getting powerful someday.
For now it is a research paper, not a lab roadmap, but coming from inside the field that would trigger the explosion, arXiv:2609.36054 is worth reading as an early warning rather than a hypothetical.