AP survey finds AI labs eye recursive self improvement soon
TL;DR
An Associated Press explainer published September 19 says leading AI developers now see so called recursive self improvement, where models help design and train more capable successors, as a near term scenario. Researchers interviewed by AP describe diverging timelines and risk estimates but broadly agree that current agentic workflows are an early form of systems doing more of their own R and D.
About this summary
This article aggregates reporting from 1 news source. The TL;DR is AI-generated from original reporting. Race to AGI's analysis provides editorial context on implications for AGI development.
Read next
Race to AGI Analysis
Recursive self improvement is the classic AGI nightmare scenario, and AP putting it on the wire signals that this is no longer a fringe thought experiment. Labs are already using swarms of agents to run experiments, optimize model architectures and even propose new training runs. In that sense, a primitive form of “AI doing AI research” is here; the debate is about when that loop becomes tight and general enough that humans lose meaningful control over the direction of improvement. ([apnews.com](https://apnews.com/article/1526da03842cfeef12d0fb69b6b7ad28?utm_source=openai))
For Race to AGI readers, the important shift is that mainstream developers are starting to acknowledge RSI as a concrete design consideration rather than speculative sci fi. That will influence everything from how compute budgets are allocated to how regulators think about cap thresholds and kill‑switch requirements. If you believe RSI is close, you prioritize interpretability, evals and containment; if you think it is decades away, you keep optimizing for speed and cost. This split is now emerging inside labs themselves, which could create internal factions over how aggressively to push the next generation of models.
