AI
Claude Now Leads a Quarter of the Research Building Its Successor, and Anthropic Published the Receipts
On Thursday, Anthropic revealed something the industry has whispered about for a year: Claude now leads 26 percent of the artificial intelligence research and development work inside the company building its successor. That word, leads, comes from a new public yardstick, the Anthropic R and D Automation Index, built with the independent nonprofit Epoch AI. At that level, Claude completes most of a task end to end from a high level prompt while a human supervises. The company will publish the figures regularly so everyone outside the lab can watch the curve.
The curve is the story. In March, Claude led 1 percent of the work. By August, the figure stood at 26 percent. More than 90 percent of the company's research now happens in collaboration with the model, and about 30,000 AI agents were doing research and engineering on Anthropic's main internal platform at any one time in August. Every one of those agents' actions gets screened before it runs, and of more than a billion decisions in August, roughly one in 47,000 was blocked. This is supervised autonomy at massive scale, with human review gating every single action.
The deeper play is transparency as strategy. Anthropic framed the release as a public good, arguing that minimizing the gap between what frontier labs know and what the public knows gives society a real say in how fast this goes. It explicitly invited other labs to publish the same metrics on the same methodology so the numbers can be compared across companies over time. That is a subtle gauntlet: once one lab shows its speedometer, every lab racing quietly gets asked why its own dashboard stays dark.
And it landed the day after OpenAI said it would regularly publish reports of unexpected model behavior, disclosing six fresh cases. Read the two announcements together and the industry's safety conversation just shifted from private assurances to public scoreboards. The labs are now competing on openness, and that competition favors whoever moves first.
Anthropic also disclosed that about 6 percent of its AI research compute went to safety work in a sample week in July, rising to 12 percent for the research carried out by AI itself. Those figures are conservative, since compute that advances safety and capability equally was counted as capability. The company calls recursive self improvement the scenario everyone needs to measure: a model that can build its own successor with growing independence. Its index sits at AL4, leads. AL5, full autonomy with zero human involvement, remains unreached in every measured area.
What this means for you: the development loop behind the most advanced AI systems is accelerating while humans still hold the steering wheel, and for the first time you get to watch the odometer. The next decade of this technology will be shaped by how quickly that 26 percent grows, and Anthropic just made that number a public fact everyone can argue about. Follow the index, and you will see the future arriving well before the press releases do.
Quick answers
What is this story about?
On Thursday, Anthropic revealed something the industry has whispered about for a year: Claude now leads 26 percent of the artificial intelligence research and development work inside the company building its successor. That word, leads, comes from a new public yardstick, the Anthropic R and D Automation Index, built with the independent nonprofit Epoch AI. At that level, Claude completes most of a task end to end from a high level prompt while a human supervises. The company will publish the figures regularly so everyone outside the lab can watch the curve.
Why does this story matter?
What this means for you: the development loop behind the most advanced AI systems is accelerating while humans still hold the steering wheel, and for the first time you get to watch the odometer. The next decade of this technology will be shaped by how quickly that 26 percent grows, and Anthropic just made that number a public fact everyone can argue about. Follow the index, and you will see the future arriving well before the press releases do.
Sources
- Reuters: Anthropic says Claude now leads a quarter of work building its next AI models
- TechXplore: Anthropic says its model Claude is helping build the next version of itself
- Digit.in: Anthropic says Claude is helping build future AI models
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.