What happened
Governance AI (Oxford) published a research paper on the potential for automating AI R&D to trigger an 'intelligence explosion,' compressing 'years of AI progress into months or less.' It cites accelerating evidence: 'As of August 2026, Anthropic reported AI systems completing 26% of internal AI R&D work with only high-level supervision, compared to just 1% five months earlier,' and OpenAI similarly reporting systems completing multi-day R&D tasks. The paper argues such an explosion could bring extreme benefits but extreme risks — AI-enabled pandemics, cyber attacks on critical infrastructure, labour disruption, and systems escaping human control — and urges policymakers to prioritise visibility into R&D automation (embedded auditors, indicator reporting), mechanisms to steer and constrain acceleration (speed limits, air-gapped secure environments, incident sharing, agreements), and societal preparedness.
Why it matters
One of the clearest current articulations of the automated-AI-R&D / 'intelligence explosion' risk and a concrete oversight agenda for governments — material for any leadership assessing existential-risk claims and R&D-governance policy.
Action needed
Familiarise policy and risk leadership with the visibility/steering/preparedness recommendations and factor R&D-automation indicators into situational awareness.