Cognizant opened its healthcare platform to AI agents, starting with the approval paperwork that slows care down.
TriZetto Unify now treats software agents as first-class users of prior authorization, under policy controls and a full audit trail, on the standard language for health data.
On May 29, 2026, Cognizant did something more concrete than most of the AI governance talk that surrounds it: it shipped. The company opened TriZetto Unify, the software that runs behind a large share of US health insurance operations, so that AI agents can use it directly as first-class users, on the same footing as a human employee. The first job handed to them is electronic prior authorization, the approval step that decides whether a treatment gets a green light, and the part of healthcare famous for making patients wait.
The scale is not small. Cognizant says the TriZetto platforms touch more than 200 million health plan members in the United States and process over $500 billion in annual healthcare spending. Letting agents into a workflow that large is exactly where an ungoverned AI becomes a lawsuit, so Cognizant built the entry the careful way. The agents operate under policy controls, every action is recorded for review, and the connections are built on the accepted industry standard for exchanging health data, so the system speaks the language that regulators and hospitals already expect. Prasad Sankaran, who runs Cognizant's AI products, called it the first move in opening the company's healthcare platforms to a new kind of consumer.
The design choice worth noticing is what stays human. Cognizant's pitch is to take administrative cost out of the system without touching clinical judgment. The agents handle the mechanical first pass at machine speed: is an authorization required, what documents are needed, submit the request. The cases that need a doctor still go to a doctor. That is the whole argument for governed autonomy in one workflow. The agent is trusted with the volume, not the verdict, and the audit trail proves which was which.
This is the most operationally specific proof in the field right now, and it doubles as a land grab. If TriZetto becomes the reference example for how regulated AI is supposed to be deployed, the standard for the whole category gets set inside one company's platform. The unanswered question is ownership. A health plan running on this gets a governed workflow, but it is Cognizant's governed workflow, defined and maintained by Cognizant. The control layer is real. Whether the client owns it or rents it is the part the announcement does not say.