← Latest AI News

November 7, 2026 · Daily brief

AI labs just doubled their speed—here's what to watch

Sovereignty angle
AI that builds better AI is the loop every lab wants. It is now running inside Anthropic, and it means the tools you see in January may be sharply better than what you are using today. Plan for that.

An independent review found Anthropic is building better AI models twice as fast as last year, mostly because AI itself is helping. No clear slowdown in sight.

What changed

On November 5, METR—an independent AI safety nonprofit—published an assessment of Anthropic's internal progress after 11 weeks of access to the company's data and staff. METR estimates Anthropic's AI models are improving on well-defined tasks at roughly 1.9 times the pace seen from March 2024 to October 2025, with a confidence range of 1.3x to 3x. METR attributes around 80 percent of the speedup to Anthropic using its own AI to help with research and development—writing code, analyzing experiments, and speeding up tasks that used to take researchers hours. The acceleration is not coming from more computing power; it is coming from smarter use of the AI that already exists. Anthropic redacted most of the detailed evidence from METR's report, so the public cannot verify the full findings, but both METR and Anthropic confirmed the overall conclusion.

What you can do with it this week

If you are testing AI tools or planning to adopt them in the next six months, expect the options available in early 2027 to be noticeably better than what you see today. That means two things: first, do not over-invest in a single tool right now if you can wait a few months. Pick monthly plans, not annual contracts. Second, set a calendar reminder for January to re-test the tasks you tried with AI in 2026 and saw fail or deliver mediocre results. The models improving faster means your team may be able to hand off work in Q1 2027 that was not ready in Q4 2026. If you are already using AI, check quarterly whether the same task now costs less or works better on a newer model. The faster pace means checking every three months instead of every six.

What to keep in your own hands

AI helping build AI is the loop that worries safety researchers, and METR found no clear bottleneck that would slow it down. That does not mean runaway progress, but it does mean you should not assume the pace stays constant. Every time you adopt a new AI tool, check whether you can export your data, switch providers, or turn it off without losing your work. Avoid tools that lock your customer list, your pricing, or your knowledge base inside a system you cannot leave. And keep a human in the loop on any task where a mistake would cost you a customer, a contract, or a compliance penalty. The models are getting better, but the faster they improve, the less time you have to test them before the next version arrives. Speed is useful only if you can still check the work.