Overview

Within this time window, two reports bring AI safety back to one question: systems are taking on a growing share of research work, but whether this could trigger an “intelligence explosion” remains a scenario, not an established outcome. Another public discussion offers a reminder that taking risk seriously need not mean giving in to panic.

The report describes a possible feedback loop

A report co-authored by more than 20 researchers defines an “intelligence explosion” as a possible cycle in which AI contributes to research and development, helps build more capable successors, and those successors take on an increasing share of the same work—potentially compressing years of progress into months.

Its authors include Geoffrey Hinton, Yoshua Bengio and researchers from organizations including OpenAI, Anthropic and Microsoft. They call for greater visibility into automated AI R&D, along with independent audits, ways to slow development and emergency plans.

Automation is growing, but metrics are not a threshold

Citing company-reported measures, the report says AI-generated code accounted for more than 80% of Anthropic’s approved code by May 2026; another measure put the share of R&D work completed with only high-level human supervision at 26% in August, up from 1% in March. These figures show AI taking a larger role in research, but they do not prove that recursive self-improvement has begun.

The authors explicitly describe the evidence as preliminary, mixed and partly indirect. They also say that productivity gains from automated AI R&D have not yet reached the threshold needed to trigger an intelligence explosion.

Warnings are more useful when paired with workable governance

Another report in the same window records Pope Leo XIV saying that concerns about AI going rogue should not be dismissed as fake news and that experts’ safety concerns deserve discussion and action; he also said he was not in a state of panic. That distinction matters: acknowledging risk is not the same as claiming the worst case is imminent.

The report’s proposals—transparent progress reporting, independent audits, limits on the pace of R&D and advance preparation for emergencies—turn the debate into governance tasks that can be examined one by one. Whether they work will depend on who verifies the metrics, how pause conditions are set and how quickly the technology changes.

We cannot yet say whether or when it will happen

The report also identifies factors that could slow a feedback loop, including limits on compute and data, diminishing returns, tasks that are difficult to automate and long training runs; AI could also help reduce risks more quickly. Whether this acceleration will occur, when it might occur and how it would affect everyday life remain uncertain.

What may be more useful to watch than a dramatic countdown is which research tasks AI actually performs, whether those changes can be independently observed, and whether society preserves room to pause and adjust when the evidence is incomplete.

Sources for this reading

- The Guardian, published 28 Sep 2026 at 11:00 EDT, updated 15:29 EDT (Beijing time 28 Sep 23:00, updated 29 Sep 03:29): AI godfathers warn of runaway ‘intelligence explosion’

- Associated Press, published 28 Sep 2026 at 20:02:41 UTC (Beijing time 29 Sep 04:02:41): Pope Leo says artificial intelligence safety concerns should be taken seriously

- CASP working paper, cover dated Sep 2026; The Guardian reports publication on 28 Sep. The PDF does not display a more precise time: What if automating AI R&D triggers an intelligence explosion?