Anthropic publishes internal metrics on how much AI now builds its own successors

Anthropic has published a set of measurement tools intended to show the public how fast AI development is moving inside frontier labs, along with a snapshot of those metrics from its own operations. The post covers three aspects of development: the extent to which AI is building the next version of itself rather than being built by humans, Anthropic's ability to oversee and intervene in actions that AI agents take on its systems, and the resources that power the development of more capable models. For each measurement the company describes what it measured, what the result showed and what would change it. Anthropic says it would expect the numbers to shift if labs coordinated on pacing the frontier, as CEO Dario Amodei called for last week. It also says it plans to embed independent third-party evaluators from multiple organizations and give them access to internal processes, systems and data comparable to what its internal risk assessment teams have. Those evaluators would verify safety practices, report incidents and monitor metrics such as the ones in the post. The publication follows questions from outside evaluators and lawmakers about how independent embedded auditors can really be, and about whether Amodei's plan needs legal backing to work.