Assessing AI risks in the context of their use, rather than as inherent in the technologies themselves, leads to smarter and safer implementations.