13 rules before an AI agent gets its own login at your practice

Most of the AI tools practices bought in the past two years write things: a draft note, a portal reply, an appeal letter. The next wave does things. Agents built into EHRs, revenue cycle platforms and scheduling systems now check eligibility, submit prior authorizations, pull charts and book visits without a person clicking through each step. Imprivata commissioned a survey of 250 U.S. health care leaders. It found 28 percent of organizations already have agentic AI in production and another 44 percent are piloting it. In the same survey, 72 percent said AI tools or agents go live without formal IT approval at least some of the time.

That gap matters more for an agent than for a chatbot. When a chatbot gets something wrong, you get a bad paragraph. When an agent gets something wrong, or gets tricked, it acts on the mistake with whatever access it was given. HIPAA doesn't exempt software. The technical safeguards in 45 CFR 164.312 still require unique user identification and audit controls for anything that touches electronic protected health information. The AMA's Oct. 13 governance framework for agentic AI also asks the question every practice will face sooner or later: who is accountable when the agent acts?

Few practices will build an agent. Most will switch one on inside software they already pay for, so the rules set at the start are the main control a practice has. Here are 13 to put in place before an agent gets its own login.