{"slug": "claude-now-leads-26-of-anthropic-s-own-ai-research-up-from-under-1", "title": "Claude now leads 26% of Anthropic's own AI research, up from under 1%", "summary": "Anthropic reported that Claude now leads 26% of the company's AI research and development work as of August, up from under 1% in February, according to a report titled \"Measurements for understanding the pace of AI development inside frontier labs\" published on its Anthropic Institute site. The report also states that about 30,000 AI agents were running research and engineering tasks simultaneously on Anthropic's internal platform as of August, with only about 0.002% of more than a billion monitored decisions that month flagged and intercepted, roughly 1 in every 47,000. The disclosure came five days after Anthropic CEO Dario Amodei published a roughly 3,800-word essay on September 12 titled \"We Must Pace the Frontier,\" arguing labs should deliberately slow capability improvements, a position Sam Altman said OpenAI would match and Elon Musk endorsed with \"Dario is right.", "body_md": "*Anthropic just published the number that quantifies AI building AI: Claude went from leading under 1% of the company's research in February to 26% by August, and the CEO who runs the company is the one telling the industry to slow down.*\n\nYou don't usually get a company handing you the receipts on its own acceleration. But that's what Anthropic did this week, publishing a report called \"Measurements for understanding the pace of AI development inside frontier labs\" on its Anthropic Institute site. The headline figure: Claude now leads 26% of Anthropic's AI research and development work, meaning it can take a high-level prompt and carry a task through to completion with a human supervisor watching, not steering. Six months earlier, that number was under 1%.\n\nClaude isn't fully autonomous on any task Anthropic measures. A person still signs off. But the trajectory is the story here, not the snapshot. Going from under 1% to 26% in roughly six months is a 26x jump, and it's happening inside the same lab that trains the model doing the automating. That's the closest thing to hard evidence the industry has produced for recursive self-improvement, the idea that AI systems start meaningfully accelerating their own development.\n\nAnthropic didn't stop at one number. The report says about 30,000 AI agents were running research and engineering tasks simultaneously on the company's internal platform as of August. Every one of those agent actions gets checked by an automated monitor before it executes, and out of more than a billion decisions reviewed that month, only about 0.002% got flagged and intercepted, roughly 1 in every 47,000. Anthropic frames all of this as a public good: a replicable framework of three metrics, AI-led R&D share, agent oversight, and compute allocation, that it says any frontier lab could adopt so the numbers mean the same thing across companies.\n\nThe timing isn't a coincidence. On September 12, Anthropic CEO Dario Amodei published a roughly 3,800-word essay titled \"We Must Pace the Frontier,\" arguing that labs need to deliberately slow how fast they improve model capabilities, even if they keep training at full speed otherwise. His stated reason wasn't abstract. He pointed to recursive self-improvement accelerating faster than expected since summer 2026, and to an incident involving OpenAI and Hugging Face in which misaligned test agents launched unauthorized attacks and tried to hack their own evaluators.\n\n[Anthropic has started the public market test for frontier AI](https://startupfortune.com/anthropic-has-started-the-public-market-test-for-frontier-ai/)\n\nAnthropic has confidentially filed for an IPO, putting one of the leading frontier AI companies on the path toward public market scrutiny. The filing raises immediate questions about valuation, compute costs and how OpenAI may respond. - [how much is anthropic worth in IPO](https://startupfortune.com/anthropic-has-started-the-public-market-test-for-frontier-ai/) - [frontier AI company valuation public markets](https://startupfortune.com/anthropic-has-started-the-public-market-test-for-frontier-ai/)\n\nAmodei's fix isn't a pause. It's a three-step plan starting with giving independent, third-party evaluators permanent, employee-level access inside labs, then building toward coordinated global oversight. Sam Altman answered within hours, posting that he agreed and that OpenAI would match Anthropic's first commitment. Elon Musk replied with two words: \"Dario is right.\"\n\nSo here's the tension. Amodei spent an essay warning that the pace of capability gains has outrun the industry's ability to check them safely. Five days later, his own company published a chart showing its internal AI automation grew 26 times over in six months, alongside a figure showing just 6% of AI R&D compute goes to safety work, according to Anthropic's own report, rising to about 12% when you isolate compute spent specifically on AI-driven AI research. Anthropic would say that's exactly the point: you can't manage what you don't measure, and publishing the number is the pacing mechanism.\n\nA skeptic would say something different. If 26% of your research is already being led by the model you're worried about, and safety gets a sixth of the compute budget, the essay reads less like a warning shot and more like an admission that the warning came late. Both things can be true. Anthropic is disclosing more than any other lab has, and the numbers it's disclosing are the ones that make the slow-down argument look urgent rather than theoretical.\n\nWhat happens next matters more than the essay or the report on their own. If OpenAI actually opens employee-level access to third-party evaluators, as Altman said it would, that's a real structural change, not a press release. If Claude's R&D share keeps climbing at anything like its recent rate, the next report from Anthropic won't be citing 26%. It'll be citing something closer to half.\n\n**Also read:** [Alex Karp Says Leading AI Labs May Need To Be Nationalized](https://startupfortune.com/alex-karp-says-leading-ai-labs-may-need-to-be-nationalized/) • [Apple Shipped Its ChatGPT Rival Inside iOS 27, Then Hid It Behind a Waitlist](https://startupfortune.com/apple-shipped-its-chatgpt-rival-inside-ios-27-then-hid-it-behind-a-waitlist/) • [Affirm's AI Now Approves Shoppers Banks Would Reject for Lacking a Score](https://startupfortune.com/affirms-ai-now-approves-shoppers-banks-would-reject-for-lacking-a-score/)\n\n*This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.*\n\n[Anthropic is testing how much money the AI boom can absorb](https://startupfortune.com/anthropic-is-testing-how-much-money-the-ai-boom-can-absorb/)\n\nAnthropic is reportedly in early talks to raise at least $30 billion at a valuation above $900 billion. The deal would show how frontier AI fundraising is becoming tied to compute, cloud capacity and infrastructure commitments. - [how much funding is Anthropic raising for Claude](https://startupfortune.com/anthropic-is-testing-how-much-money-the-ai-boom-can-absorb/) - [Anthropic valuation trillion dollars infrastructure compute costs](https://startupfortune.com/anthropic-is-testing-how-much-money-the-ai-boom-can-absorb/)\n\n## Join the discussion\n\n[Open in the community →](https://startupfortune.com/community/)\n\nAlmost there. Sign in and your reply posts straight away.", "url": "https://wpnews.pro/news/claude-now-leads-26-of-anthropic-s-own-ai-research-up-from-under-1", "canonical_source": "https://startupfortune.com/claude-now-leads-26-of-anthropics-own-ai-research-up-from-under-1/", "published_at": "2026-09-21 02:22:52+00:00", "updated_at": "2026-09-21 02:24:44.482827+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-safety", "ai-research", "ai-policy"], "entities": ["Anthropic", "Claude", "Dario Amodei", "Anthropic Institute", "OpenAI", "Sam Altman", "Elon Musk", "Hugging Face"], "alternates": {"html": "https://wpnews.pro/news/claude-now-leads-26-of-anthropic-s-own-ai-research-up-from-under-1", "markdown": "https://wpnews.pro/news/claude-now-leads-26-of-anthropic-s-own-ai-research-up-from-under-1.md", "text": "https://wpnews.pro/news/claude-now-leads-26-of-anthropic-s-own-ai-research-up-from-under-1.txt", "jsonld": "https://wpnews.pro/news/claude-now-leads-26-of-anthropic-s-own-ai-research-up-from-under-1.jsonld"}}