The market asks whether, by the end of 2030, an AI will publicly demonstrate the complete autonomous R&D cycle needed to create a more generally capable version of itself, with credible external recognition. Current systems show meaningful progress in automating agent design and model post-training, but the reported gains remain bounded or harness-level; they do not yet establish autonomous creation of a broadly more capable successor, and evaluations still identify gaps in open-ended research autonomy. Four years leaves substantial time for improvement, so the evidence supports a moderate chance rather than treating current limitations as decisive. External sentiment found markets about an announcement by OpenAI or Anthropic in 2026, not this exact 2030 outcome, so those short-horizon probabilities are not a direct anchor.
Recursive says its Eureka Machine contributed across development of an open-weight 40B coding model. Coverage describes substantial human oversight and no demonstrated full recursive self-improvement, making this evidence of R&D automation rather than the strict endpoint.
TIME reports that interviews with researchers at leading labs describe efforts to automate ideation, experiments, and interpretation. Forecasts that RSI could begin within 3–12 months are attributed expectations, not public verification of full RSI.
Clark cites reported progress on TasteVal, but the benchmark covers experiment design and interpretation on fixed problems rather than the complete autonomous research-to-successor cycle; this is a forecast, not evidence that the market's qualifying event occurred.
A September technical report describes an eight-day run in which the agent rewrote its own harness and selected successive improvements, with reported transfer to held-out tasks. This is a bounded agent-layer result, not evidence that a system autonomously directs and completes its own general-capability R&D.
A modeling paper distinguishes narrow AI-R&D benchmark gains from broader capability improvements and estimates current feedback loops are not yet strong enough for self-sustaining acceleration, though they are strengthening. Harvard Gazette coverage also discusses concerns about defining a genuinely more capable AI successor.
Will superintelligent software attack people at will by 2036?
Will AGI not yet be developed by the end of 2030?
Will AGI be developed and be net positive to the world by the end of 2030?
Will super intelligent software extort people at will to gain advantage within the next 5 years?
Will AGI be developed and be net negative to the world by the end of 2030?
Will any meaningful federal AI regulation be enacted into law by 2029?
