Anthropic publishes three internal measures of AI development
The company says it will track how much of its own research is done by AI, how well agents are overseen, and how compute is allocated.
Anthropic said on Thursday it will publish three measurements intended to show the public how AI development is progressing inside the company. They cover how much of its AI research is carried out by AI, how well AI agents are supervised, and how its compute is allocated.
The company framed the disclosure as a response to systems being used to build their own successors. It published a snapshot of current values alongside the announcement.
The measures are self-reported. Anthropic defines each one, collects the underlying data and decides what to publish, so the figures describe what the company chooses to count rather than an externally verified position.
That said, the categories are the ones outside observers have asked about. A number for how much research is automated is a claim a company can be held to over time, which is more than most laboratories currently offer.
No other frontier laboratory publishes comparable figures, so there is nothing yet to compare them against.