Anthropic shares three measures to track AI development

@AnthropicAIAI systems are getting more powerful, and they're increasingly being used to build the next version of themselves. We want to illuminate that progress for the public. Today, we're sharing three measurements that help track AI development: 1. How much AI R&D is done by AI. 2. How well AI agents are overseen. 3. How compute is allocated. We provide a snapshot of these metrics from inside Anthropic. Any frontier developer could publish the same measures, and third parties could verify them. As the world considers pacing the frontier, we should do everything possible to minimize the gap between what frontier labs know and what the public knows. This means better measuring the development of AI, publishing our findings, and giving society an opportunity to decide how to use this information. Read the full post and methodology: https://www.anthropic.com/institute/measuring-pace-of-ai-development

@natolambertI don't think this really answers the question of what is happening in "RSI". Says Claude “leads” 26% of Anthropic’s AI R&D work, but doesn't define what falls in scope of AI R&D work. I'm not satisfied by this, but it's a step in the right transparency direction.

@natolambertSome info in the appendix, but feels like R&D will include a lot of softer skills here than someone like me who's built a bunch of models would include.

@eliebakouch@AnthropicAI really great that you publish this congrats! would be great to publish the breakdown of AI usage (and automation) per research task! 🙏

@eliebakouchinteresting that openai gives us the breakdown of usage per R&D task, and anthropic gives us automation levels (both from the same @EpochAIResearch article!!) would be great to combine both and get the evolution of automation per R&D task! the compute allocation part is also very nice, only 6% of research compute went to ai safety, with 12% from the "AI driven" category!

@teortaxesTexCountry of geniuses in a datacenter.

@kimmonismusIn around six months, the share of AI model R&D tasks at Anthropic led by Claude has risen from 1% to 26%. Speaking about RSI, Claude also collaborates on or leads more than 90% of Anthropic’s model R&D work. Around 30,000 AI agents are working on research and engineering at any given time on its main internal agent platform.

@rohanpaul_aiInside Anthropic 26% of its AI R&D work has reached the point where Claude can do most of the task end-to-end from a high-level instruction. A human mainly supervises. - “As of Aug 2026, there were approximately 30,000 agents doing research and engineering work at Anthropic at any one time in our most-used internal platform.” - More than 90% of measured Anthropic R&D already has AI doing substantial chunks of the work or more “The share of work at or above “AI collaborates” is above 90%.” - More than a billion agent decisions in one month is the scale here; the online monitor blocked roughly one in every 47,000. “We analyzed over a billion decisions from our research & engineering agents over August 2026 and found that 0.002% of them (about 1 in 47,000) were blocked by this monitor.”

@shiringhaffaryIn a new post, Anthropic says Claude AI leads 26% of all its R&D development, in one of the clearest indications so far that AI is speeding up development of future models. https://www.bloomberg.com/news/articles/2026-09-17/anthropic-says-claude-drives-26-of-its-research-and-development

@shiringhaffarySome other notable facts in here: -Anthropic estimates 6% of all compute that goes twd AI R&D goes toward safety rather than building capabilities. Will be interesting if that changes. -Anthropic has ~30,000 agents doing internal research and engineering work

@JeffLadishGreat to see this from Anthropic. A few weeks ago I wrote that the company could be a lot more transparent, and I appreciate how they’ve stepped up, both with this and the recent incident investigation report and misuse report

@BorisMPowerWow, congrats to Anthropic at quantifying, being transparent and investing effectively to be in this moment in time!

@jackclarkSF@natolambert Thanks for your thoughts - this was a project I helped supervise as a follow-on to When AI Builds Itself. We're sharing this as a way to put more information and methodology for measuring RSI into the world. I agree with you the definitional aspects of this are v complicated

@eli_liflandRT @ChrisGPT: Wow Anthropic is being really transparent with how close to RSI they are - 26% of Anthropic’s AI R&D work,” while “the share…

@ziv_ravidClaude "collaborates" on 90% of Anthropic's research, defined as doing "large chunks under close human direction." By that definition, autocomplete collaborates on 100% of my papers (pre llms ofcourse)

Anthropic shares three measures to track AI development

TLDR

Combined views

Anthropic shares three measures to track AI development

15 Sources

Combined views

Sentiment

Summary

TLDR

Sentiment

Summary

Sentiment

Summary

15 Sources

Sentiment

Summary