What it is. AI R&D acceleration means AI tools helping researchers build better AI faster: writing code, running experiments and analysing results that people used to do by hand.
An analogy. Imagine a bakery where the bakers build a better oven, and the better oven lets them bake the next oven faster. Each round goes quicker than the last. That feedback loop is what researchers mean when they talk about AI speeding up its own development.
Why safety researchers watch it. If each generation of AI helps build the next one sooner, progress could speed up faster than companies, regulators and testers can keep up with. It is one of the scenarios safety researchers and governments watch most closely, which is why outside measurement matters.
Why it showed up in the news this week. On 5 November METR, an independent AI evaluation group, published a public summary of an assessment it carried out with high access inside one frontier lab between 12 August and 29 October. On its own benchmark trend, the lab’s models have improved since October 2025 at about 1.9 times the earlier rate. METR estimates that about 80% of the acceleration is due to AI, with substantial uncertainty, and is confident it was not driven by faster growth in training compute. Our brief has the details.
The caveats. The lab could redact the report, and removed much of the evidence METR found most informative, so readers rely on METR’s judgement. The gains may also be limited to tasks that are easy to grade automatically, which is not the same as all of AI research.
One thing to remember. The question is not just how good AI is, but how fast it is getting better. Outside checks like this one are the first real attempts to measure the second question.
Source: METR, assessment summary, 5 November 2026, metr.org.