AI Is Dramatically Accelerating AI Research, METR Report Finds

Anthropic defines the threshold in its Responsible Scaling Policy, effective July 2026, as the point where overall capabilities progress is observed or expected to run twice as fast as the fastest sustained rate seen before AI contributed much to AI R&D. The company says that pace must be plausibly attributable to AI automating AI R&D, rather than to faster compute or other inputs.

Cotra, who did not work on the project and did not see the unredacted report, says the acceleration may have started for a mix of reasons. Still, she concludes that AI automation is probably the large majority of the explanation.

The core concern is compounding. If each generation of models shortens the time needed to build the next one, the curve steepens rather than flattening, which is why the "superintelligence within a year" extrapolation draws so much attention even as researchers caution that it is naive.

The report also raises questions about transparency. With the most informative evidence redacted, outside observers cannot independently verify the pace or the cause, leaving the public to weigh a summary against Anthropic's own policy language.

For readers in Tunisia, the practical takeaway is not a product launch or a price. No availability, pricing or release details are attached to this research, and nothing has been announced for the local market.

What is worth watching is the direction of travel. If frontier labs keep accelerating their own research, the models that reach Tunisian developers, startups and students through APIs and cloud services could improve faster than the usual yearly upgrade cycle. For now, the acceleration claim rests on an internal metric and a partly redacted report, so it should be treated as a strong signal rather than a settled fact.