System prompt
A fixed set of instructions that shapes how an AI model behaves before the user types anything. It sets the role, the rules, the tone, and often the safety limits the model should follow in every conversation. Two organizations using the same underlying model can get very different behavior because their system prompts differ. For people accountable for a deployed tool this matters: a lot of what makes a healthcare AI product safe or unsafe is decided in the system prompt and the guardrails, not in the model itself, so it is fair to ask to see it.
A patient portal assistant runs on a general-purpose model, and its system prompt is what tells it to direct anyone describing chest pain or stroke symptoms to emergency services and to decline questions about doses. When the vendor later rewrites that prompt to make answers shorter, the escalation instruction gets trimmed along with everything else, and no one at the hospital hears about it. Put in the contract that you can see the system prompt and get notice before it changes, and rerun your own set of red-flag test messages after every change.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →