Anthropic Weighs Human Review for Some R&D Decisions as Models Take On More Work

The signal here is governance keeping pace with automation inside a lab whose output shapes the broader AI market, not a product change. Anthropic acknowledging that models now carry a meaningful share of R&D work implies internal decision rights are shifting toward systems — and that oversight is being retrofitted rather than designed in. The absence of specifics on which decisions get human sign-off, or timing, is the substantive gap: scope defines whether this is a narrow checkpoint or a real constraint on autonomy. How other frontier labs describe comparable internal controls is worth watching.