First Public Glimpse of an In-House Automation Gauge

The 26 percent figure comes from an R&D metric the company had not previously published, marking the first time the lab has quantified in public how much of its model-development workload is carried out by AI systems rather than human engineers. The claim circulated in Korean coverage under the headline framing "AI now makes 26 percent of AI," prompting a review against three primary sources, which confirmed the underlying disclosure and the figure as stated by the company.

출처: 조사 출처

What the 26 Percent Actually Measures

The number reflects AI's share of internal development tasks under Anthropic's own definition of R&D work — not a claim that its models design, train and ship AI entirely without human involvement. Headline framings that suggest fully autonomous AI research stretch beyond what the indicator supports, and because the gauge is defined in-house, it offers limited basis for comparison across rival labs. For semiconductor and hardware watchers, the disclosure still lands with weight: a rising AI share in AI development is a direct signal of growing compute intensity inside the labs that buy the most advanced accelerators.

Core Disclosure Holds, Framing Carries Caveats

The essential elements of the claim — that Anthropic disclosed an internal R&D metric for the first time and put AI's contribution to its AI development at 26 percent — hold up against the primary record, while looser interpretations of the figure overstate its scope. On that balance, the claim lands as Mostly True.