What happened
Governance.ai argues that AI is automating its own R&D fast enough to make an 'intelligence explosion' a scenario leaders must plan for: as of August 2026 Anthropic reported AI completing 26% of internal AI R&D work with only high-level supervision (up from 1% five months earlier), and OpenAI reported systems 'routinely completing R&D tasks that would take days for staff.' The paper warns such acceleration could compress years of progress into months, 'be the most consequential technological development in history,' and outpace society's ability to absorb risks like AI-enabled pandemics, cyber attacks on critical infrastructure, and loss of human oversight. It sets out three prioritized policy moves: visibility into R&D automation via embedded auditors and reporting, steer-and-constrain tools (speed limits, air-gapped R&D environments, verification agreements), and societal preparation including emergency response plans.
Why it matters
This gives boards and policy leads a concrete, dated early-warning dataset on AI-automated R&D (26% of frontier R&D now AI-performed) and a ready-made vocabulary for debating 'decisive AI lead' scenarios in strategy and risk conversations.
Action needed
Brief senior leadership on the R&D-automation indicators (Anthropic 26%, OpenAI days-long tasks) and evaluate whether your own AI strategy assumes a pace of capability growth that these numbers call into question.