22 Researchers Including Hinton and Bengio Warn AI Could Automate Its Way to an Intelligence Explosion

Cambridge's Center for the Study of Existential Risk-adjacent program CASP (AI Science and Policy) published a September paper, "What If Automating AI R&D Triggers an Intelligence Explosion?," signed by 22 researchers including Geoffrey Hinton, Yoshua Bengio, OpenAI chief scientist Jakub Pachocki, Anthropic co-founder Jack Clark and Meta researchers. It examines recursive self-improvement (RSI): whether handing AI R&D work to AI itself could suddenly accelerate progress. Hinton tweeted on Oct 3 that many top researchers now believe this could happen in the near future.

Confirmed

Hinton tweeted on Oct 3 that the idea of an RSI-driven intelligence explosion is longstanding but only recently has become imminent, linking to the paper he co-authored.
The report was released in September via Cambridge CASP with 22 signatories, including Alan Chan as co-author.
The report argues AI is already automating AI R&D and an intelligence explosion may be closer than assumed; per 新智元, the time window for humanity could be only months.
Cited real-world evidence includes, per APPSO, Anthropic's internal share of AI-generated approved code rising to 80% since 2025; 量子位 reported RSI could accelerate technical progress tenfold within a year.

Why it matters

The paper's rare convergence of core researchers from OpenAI, Anthropic and Meta alongside academic leaders like Hinton and Bengio signals rising mainstream consensus on intelligence-explosion risk.
The 80% AI-generated code figure provides quantifiable evidence that AI is deeply involved in AI development, grounding the debate beyond theory.
If RSI truly accelerates progress, the window for AI safety and governance may be far shorter than previously expected.

2026-10-03 ~ 2026-10-05 · 7 related posts

Primary sources

Hinton: leading researchers now think an intelligence explosion may happen soon — geoffreyhinton · 2026-10-03 04:42 (UTC+08:00)
Hinton, Bengio and 20 top researchers warn an AI intelligence explosion may be months away — 新智元 · 2026-10-03 12:19 (UTC+08:00)
Cambridge report: AI-written code hits 80% at Anthropic as RSI could speed progress 10x — 量子位 · 2026-10-03 14:00 (UTC+08:00)

[source] Hinton: leading researchers now think an intelligence explosion may happen soon — geoffreyhinton · 2026-10-03
Hinton, Bengio and 20+ researchers warn AI-run AI R&D could trigger intelligence explosion — tinyfool · 2026-10-03
[source] Hinton, Bengio and 20 top researchers warn an AI intelligence explosion may be months away — 新智元 · 2026-10-03
Hinton, OpenAI and Anthropic researchers unpack recursive self-improvement in new report — APPSO · 2026-10-03
Hinton, Bengio join paper warning AI R&D automation could trigger an intelligence explosion — LuizaJarovsky · 2026-10-05

2 near-duplicate retellings: 量子位 · LuizaJarovsky