Your AI avatar can speak for your company. Who approves the script?

AI avatars are becoming more common

Companies can now create new videos from an employee’s likeness and cloned voice without another recording session. Tools such as Visla’s AI Avatar creator also let teams build synthetic presenters or choose from a public avatar library.

That makes it easy for one project to spread. An employee creates an avatar for internal training. Marketing later wants it for a product launch, an agency gets access, and the script is translated. Then the employee changes roles or leaves.

The company now needs to know whether external marketing was approved, who can produce new videos, who reviews translations, and whether old videos can stay online. Businesses should answer those questions before the first project becomes a shared resource.

Give every AI avatar a clear owner

Every company avatar should have a designated owner who oversees its use. This is an operational responsibility, not necessarily a claim of legal ownership.

The policy should identify:

• Who may create or import the avatar
• Who may access it
• Who approves scripts and completed videos
• Who removes access when a project, contract, or employment relationship ends

A public avatar narrating a software tutorial doesn’t carry the same weight as a digital version of the CEO. Access to a leader’s avatar should be limited, documented, and revoked when it is no longer needed.

Decide what each avatar is allowed to say

Classifying uses in advance tells teams which projects can move quickly and which require communications, legal, HR, or subject-matter review.

• Routine uses: Product tutorials, onboarding, course introductions, training updates, and localized versions of approved material
• Uses that need additional approval: Marketing claims, financial information, policy changes, professional advice, and statements attributed to a specific employee
• Uses that may require the real person: Crisis apologies, layoff announcements, safety incidents, personal testimony, and statements accepting responsibility

A CEO’s apology implies that the CEO chose to appear and accept responsibility. A customer story implies that a real person had the experience being described. Replacing either speaker with an avatar can change how viewers interpret the message, even when the script is accurate.

A synthetic presenter can explain an established company position. It shouldn’t create firsthand experience, personal emotion, or accountability that never occurred.

Define each type of AI avatar

Different avatar types create different policy questions:

• Real-person avatar: A presenter modeled on an identifiable employee, executive, contractor, or another person. It raises questions about consent, endorsement, access, and continued use.
• Fully synthetic avatar: A generated presenter that isn’t intended to represent a specific person. It raises questions about disclosure and false representations of experience or expertise.
• Public or stock avatar: A vendor-provided presenter that may be available to many customers. It raises questions about licensing, nonexclusivity, and availability.

For a real-person avatar, permission should specify:

• Approved teams, users, and outside agencies
• Internal, external, organic, and paid uses
• Permitted languages, translations, and voice modifications
• Whether approval happens per script, completed video, or campaign
• How long permission lasts
• What happens after a role change or departure

A company can approve every video, one campaign, or a narrow category such as internal software tutorials. Advertising, translation, and external distribution can require separate approval.

California’s AB 2602 digital replica law makes certain provisions unenforceable when they permit a new digital-replica performance without reasonably specific intended uses and other requirements are met. For businesses, the practical takeaway is to document the channels, audiences, languages, and content covered by permission.

Synthetic and stock avatars also need rules. They shouldn’t pose as real customers, licensed professionals, or employees describing personal experiences. Teams should review vendor terms, nonexclusivity, and availability.

A cloned AI voice deserves the same care as appearance because it can carry identity and implied approval.

Put AI avatar approval inside the video workflow

A policy won’t help if the production process gives people no place to apply it.

1. Choose an approved avatar and use case.
2. Assign a script owner.
3. Review factual, legal, and brand claims.
4. Get the represented person’s approval when required.
5. Review the completed video.
6. Add the required disclosure.
7. Record who approved and published it.

Reviewing the script alone isn’t enough. Pronunciation, pacing, facial movement, captions, editing, and surrounding visuals can alter the final message.

Shared video workspaces and time-stamped review comments can keep decisions attached to the project instead of scattered across email, chat, and separate file versions.

What UBS’s AI avatar program shows

UBS began using avatars of research analysts in 2025 to turn written investment research into short client videos. According to the Financial Times’ report on UBS’s AI analyst avatars, analysts opt in, review scripts, and approve completed videos before distribution. UBS labels the videos as AI-generated and slowed part of the rollout after the system failed to reproduce some accents accurately.

That accent problem shows why final-video approval matters. An analyst might approve every word and still object to changes in pronunciation, tone, or professional identity.

Business Insider’s follow-up on the UBS avatar program reported that 36 analysts had volunteered by May 2025. UBS also said analysts would continue reviewing videos before they reached clients, even if more of the production process became automated.

UBS separates decisions that companies often combine. Agreeing to create an avatar isn’t the same as approving a script. Script approval doesn’t replace review of the final performance, and one approved video doesn’t authorize every later use.

A smaller business may not need a bank’s review structure. It can still separate consent, content review, finished-video review, and audience disclosure.

How should businesses disclose AI avatars?

A company should decide when disclosure appears, where viewers see it, and how long it remains visible. Its standard might require:

• An on-screen label when the presenter first appears
• Written disclosure in the title, caption, or description
• A persistent label for advertising or sensitive external videos

The European Commission’s Article 50 guidance says relevant EU AI Act transparency obligations apply from August 2, 2026 and cover marking and labeling certain AI-generated or manipulated content.

New York’s law on synthetic performers in advertising requires conspicuous disclosure in covered advertisements when the advertiser knows a qualifying synthetic performer is being used.

These rules apply to specific circumstances rather than every business video. An internal standard should set a baseline while allowing for additional legal requirements.

What happens when an employee leaves or an avatar is retired?

Offboarding should address more than the employee’s account. The company also needs to locate every place where the avatar, voice, source material, or completed video remains in use.

The policy should cover:

• Access held by employees, contractors, and agencies
• Templates and duplicated projects
• Translated or localized versions
• Stored recordings, photos, and cloned voices
• Existing internal and external videos
• Permission that expires or is withdrawn

Previously approved videos and future generation are separate decisions. A company might keep an approved training video online while blocking new material. If the person’s title or authority is central to the message, the existing video may also need an update.

The National Institute of Standards and Technology, or NIST, is a U.S. government agency that publishes technical and risk-management guidance. Its Generative AI Profile recommends clear roles, system inventories, document-retention policies, human oversight, and procedures for deactivating AI systems. The guidance isn’t law, but its lifecycle approach applies well to company avatars.

AI avatar policy checklist for small businesses

A smaller company doesn’t need a 40-page manual before producing its first avatar video. It does need these decisions in writing:

1. Name the policy owner: Identify who updates the rules and resolves unclear cases.
2. List approved tools and avatar types: State which platforms, likenesses, and voices employees may use.
3. Classify use cases: Define which videos are routine, which require escalation, and which aren’t allowed.
4. Record specific consent: Document the audience, channel, language, duration, and approval model for each real-person avatar.
5. Name the approvers: Identify who checks the script and who reviews the completed video.
6. Set a disclosure standard: Specify the wording, placement, and duration of AI labels.
7. Create an offboarding process: Remove access and decide what happens to templates, voices, source files, and published videos.
8. Set a review date: Revisit the policy as the company’s tools and use cases change.

Writing these rules early makes it easier to approve new uses, trace consent, remove access, and review existing videos.