Anthropic Publishes Metrics to Track Frontier AI Progress Internally

The notable shift here is disclosure design, not the automation level: a frontier lab is publishing internal progress metrics and inviting outside evaluators, which turns capability claims into something auditable rather than asserted. That matters for regulators, enterprise buyers and rival labs under similar scrutiny. The stated ceiling — heavy collaboration, no fully autonomous subsets — is the detail worth watching, since whether that boundary holds as agent counts and R&D share grow is the open question.