Responsible AI policy

• Answering the business’s customers or patients with the information the business sets up (Anthropic’s Claude models).
• Proposing the assistant’s initial setup and suggesting improvements based on the team’s corrections.
• Transcribing voice notes (Groq) and, if the business turns them on, handling voice calls (Vapi).

People must know they are talking to an AI assistant. The business must say so, for example in the greeting or in its privacy notice, and cannot set up the assistant to deny being automated. This also meets the transparency obligations of the European Union AI Act for systems that interact with people.

ISSA supports the team and does not replace it. The team sees every conversation in the inbox, can take over a chat or pause the assistant at any time and gets alerts when the assistant asks for help. Improvement suggestions only apply if a person accepts them.

The assistant makes no clinical, legal, credit or employment decisions. With patients, it never diagnoses, interprets symptoms or recommends treatments: it passes the question to the professional and, if it sounds urgent, tells them to call the clinic or emergency services.

AI can be wrong. We limit the assistant to the business’s topics, tell it to look up prices and availability with tools instead of inventing them and protect it against malicious instructions from third parties. Even so, the business must keep its information up to date and review the replies.

We do not use customer data to train AI models. Anthropic does not, by default, use data sent through its commercial API to train its models and keeps it for a limited time under its terms. An organization’s improvement suggestions are based only on its own conversations and are never shared with another.

The clinical record (notes, history, photos, consents) is never sent to the AI model. In patient follow-up, the assistant only receives the appointment details and the procedure name needed to coordinate.

A business can register its own Anthropic key; in that case, processing by Anthropic is governed by the business’s contract with Anthropic.

If you see harmful, discriminatory or repeatedly wrong replies, write to [email protected]. We review reports and adjust the assistant’s rules.