Anthropic details practical metrics to help monitor the speed of AI development
SiliconANGLE Mike Wheatley
Anthropic says it has three ways to track how fast AI is advancing. It’s a push for more public visibility after calls to slow frontier model work.
Based on reporting by SiliconANGLE, Mike Wheatley — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic is trying to turn a big debate into something measurable. Days after Dario Amodei called for leading model makers to coordinate a slowdown in frontier AI development, the company published three metrics it says can help track the pace of progress and make that debate less abstract.
The three measures cover how much AI is doing AI research, how well humans are overseeing autonomous agents, and where compute is going. Anthropic says it is already measuring each one internally, and it laid out the methods so other frontier labs can follow the same approach.
The first metric looks at AI-led R&D. Anthropic says frontier labs are increasingly using models to build more capable models, which raises the risk that people stop understanding how far the systems have advanced. Its own index shows Claude is involved in its research work, but not operating fully autonomously in any subset of that work.
The second metric focuses on oversight of AI agents. Anthropic says people are already using agents to do work on their behalf, and those agents can pass dozens of tasks to sub-agents. To keep tabs on that, it built a system that can oversee agents and intervene inside its own systems. That setup showed about 30,000 autonomous agents running in its computing environments, most of them tied to R&D.
The third measure is compute allocation. Anthropic says compute is the main fuel for AI and one of the most verifiable inputs to the research process, so watching where it goes can reveal what developers are prioritizing. Its first snapshot, covering July 13 to July 20, showed around 6% of total compute capacity going to AI safety, with another 12% going to AI-driven R&D focused on safety.
This lands in a tense moment. Amodei’s slowdown pitch has picked up support from Sam Altman, Elon Musk and Demis Hassabis, while Anthropic researcher Evan Hubinger has warned there is a greater than 10% chance the technology could kill all humans within 10 years. Anthropic says these metrics are a starting point for outside evaluators trying to judge how fast AI is really moving, and how capable it’s becoming.
My take — AI-written commentary, not fact-checked reporting
This is Anthropic doing the oldest Silicon Valley move in the book: calling for restraint, then publishing the spreadsheet. That’s not hypocrisy; it’s strategy, and probably the only version of “safety” anyone will actually read. If frontier AI is going to be paced, someone has to measure the race instead of just talking about the finish line.
Read more about this at: SiliconANGLE
