Editorial illustration for New report on Anthropic automating AI development alarms researchers
Anthropic's AI Models Now Build Next-Gen AI, Report Shows
New report on Anthropic automating AI development alarms researchers
Anthropic's own models are now doing a sizable chunk of the work needed to build Anthropic's next models, and the pace is startling. A report published Thursday by METR, the AI safety research group, spent 11 weeks inside the company watching this play out. The conclusion: Anthropic's research is improving roughly twice as fast as it was a year ago, with no clear sign of slowing down.
The timing is awkward. Dario Amodei, Anthropic's CEO, has spent months publicly urging the industry to slow down on AI development, even as his own company's internal tools accelerate it. Anthropic is also preparing to go public, putting new scrutiny on how the company talks about risk versus how it actually operates.
The METR investigation is separate from a program Anthropic launched this summer giving outside safety researchers employee-like access to its systems, a move that followed high-profile hacking incidents. What makes this report notable isn't just what it documents about AI building AI. Anthropic was allowed to redact parts of the final version before publication, and researchers say those gaps matter.
The findings offer some of the clearest evidence yet of AI accelerating its own development, raising questions about how quickly the technology could advance toward a future in which AI systems can improve themselves with little or no human involvement. They also come at a pivotal time for Anthropic, which is preparing to go public even as its CEO, Dario Amodei, has publicly called for labs to pump the brakes on AI development.
Why this matters
Anthropic has spent the past two years warning Washington and anyone who'd listen that AI development needs guardrails, timelines, maybe a pause. Now its own internal data shows the opposite happening inside its walls: AI tools doubling the pace of AI research, with no sign of plateauing. That's not hypocrisy exactly, but it is a tell.
When Hobbhahn says the report points to R&D speeds faster than outsiders assumed, he's flagging a gap between what labs say publicly and what their engineers are doing with Claude on a Tuesday afternoon. For founders and researchers, the practical lesson is blunt: if frontier labs are already using models to design the next generation of models, the compounding curve everyone argues about in the abstract is running live at Anthropic right now. We'd treat this less as a safety footnote and more as a timeline marker.
Anyone betting on slow, linear progress to plan product roadmaps or policy responses should probably recheck that assumption against what Anthropic's own researchers are reporting about their workday.
Related Reading
Browse all Anthropic news →
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps
Related Topic Anthropic: Hackers Stole Claude Tokens Via Compromised Session Key
LLMs & Generative AI
From Archives Nine AI agents fail US national lab research bar; most skip required plans
From Archives Research: Open AI Models Now Nearly as Capable at Hacking as Top Closed Ones
From Archives OpenAI reaches USD 1.25 trillion valuation as rival Anthropic nears IPO
Business & Startups
From Archives Ramp Data Shows AI Spending Stalling as Anthropic IPO Nears
Business & Startups
Common Questions Answered
What did METR's report find about Anthropic's AI development pace?
METR's 11-week investigation found that Anthropic's research is improving roughly twice as fast as it was a year ago, with no clear sign of slowing down. The report provides evidence that Anthropic's own AI models are now doing a sizable chunk of the work needed to build the company's next models, demonstrating AI accelerating its own development.
How does Anthropic's internal AI automation contradict its public messaging?
Anthropic's CEO Dario Amodei has publicly urged the AI industry to slow down development and called for guardrails and timelines, yet the company's internal data shows AI tools doubling the pace of AI research with no sign of plateauing. This reveals a significant gap between what labs say publicly about responsible AI development and what is actually happening inside their walls.
What are the implications of AI systems automating their own development according to the report?
The findings raise questions about how quickly AI technology could advance toward a future where AI systems can improve themselves with little or no human involvement. This accelerated self-improvement trajectory suggests the industry may be moving faster toward advanced AI capabilities than outside observers previously assumed.
Why is the timing of METR's report particularly significant for Anthropic?
The report comes at a pivotal time as Anthropic is preparing to go public while its CEO has simultaneously been calling for the industry to pump the brakes on AI development. This creates a notable tension between the company's public advocacy for AI safety precautions and the evidence of its own accelerating internal development pace.
Further Reading
- Why a new report on Anthropic automating AI development is alarming researchers - The Verge
- Independent Assessment of AI R&D Acceleration at Anthropic - METR
- Review of the “Risks from automated R&D” section in Anthropic’s Risk Report - METR
- Anthropic is rapidly accelerating AI development amid slowdown calls, outside research group says - Boston 25 News
- METR finds Anthropic may have crossed its own AI acceleration warning threshold back in July - The Decoder
