Report: Anthropic's AI is accelerating AI development

A new report from METR, an AI safety research organization, finds that Anthropic is using AI to help build more powerful AI, and that the approach appears to be working. According to the findings, Anthropic's models are improving roughly twice as quickly as they were a year ago, with little sign of slowing. The result has drawn attention because it suggests AI is becoming a meaningful accelerant in AI research itself. Researchers have long flagged recursive self-improvement as a threshold to watch. The term refers to AI systems speeding up AI research in ways that could compound progress and make oversight harder. The METR report does not claim a sudden breakthrough or describe an uncontrolled system. Instead, it points to a measurable acceleration: AI tools are helping researchers do work faster, and that speed is showing up in model improvement. That distinction matters. A faster research loop is not the same as a runaway intelligence explosion, but it can still change the safety calculus. If companies can iterate more quickly, then evaluation, governance, and risk assessment need to keep pace. Otherwise, capability could advance faster than the institutions meant to monitor it. The finding also adds context to the broader competition among frontier labs, where speed is both a commercial advantage and a source of concern. For Anthropic, the report may reinforce its focus on safety research even as it pursues more capable models. For policymakers and outside researchers, the lesson is that AI-assisted AI development is no longer only a theoretical scenario. The debate is shifting from whether acceleration could happen to how quickly it is already occurring, and what safeguards should accompany it.