An executive avatar for internal updates: consent and labeling

If you make an avatar of an executive for staff updates, get written consent that covers the face and the voice, name the clips it may be used for, and label each one as synthetic. The technology can produce a convincing clip from a single photo, and that is the reason to set the rules before anyone makes the first one.

Why this is a rule now

Tavus describes its Griffin research preview as a model that works from a single reference photograph, and says it can clone a voice from about 10 seconds of audio (Tavus Griffin, read 2026-10-05). The page presents this as a research preview for selected trusted testers, not a product open to customers. The point for your policy is that both inputs are now small enough to find in a public profile, a recorded all-hands, or a conference talk. So the control cannot rely on difficulty.

A consent checklist

Sume Avatar 1.0 can create an avatar from a prompt, props, or a photo, and then reuse it by handle in later renders. A reusable handle is convenient and also the thing to govern: whoever has access can make the executive say new things.

Operating practice

• Keep a script log: script text, approver, date, and avatar_video_id.
• Have the executive approve each script. A past approval does not cover a new message.
• Never use it for announcements about layoffs, safety incidents, or anything where staff would feel deceived to learn it was synthetic.
• Use a stock or a clearly fictional host if the executive does not want to be cloned. The message can be signed by them without using their face.

Limits

Consent forms differ by country and by employment contract, so this is a checklist, not legal advice. Staff trust is the larger risk: a synthetic executive reading a warm message can feel hollow, so use it for routine updates and let the real person speak when it matters.