Responsible Scaling Policy V3

Anthropic distinguishes its own safety plans from the safeguards it recommends across the AI industry in version 3 of its Responsible Scaling Policy. Its February 24, 2026 announcement also introduces public safety roadmaps whose individual goals are explicitly nonbinding.

The accountability mechanism is disclosure: the company says it will grade progress against those goals and publish risk reports every three to six months, with external review required in specified circumstances. Those reports are intended to connect model capabilities, possible threats, and existing defenses. The policy explanation makes roadmap targets useful for tracking progress, while distinguishing them from mandatory conditions for releasing a model.