Consent and impersonation harms

Meaningful consent is informed, specific, voluntary, understandable, and revocable where possible. A person should know what parts of their identity will be captured, what will be generated, where it will appear, how long permission lasts, who can access the model or files, and what happens if they withdraw.

A broad clause hidden inside unrelated terms may satisfy neither ethical expectations nor applicable rules. Power differences also matter. An employee asked to donate a voice by a supervisor may not feel free to refuse. A child cannot be treated as a small adult, and guardianship does not erase the child’s interests and future autonomy.

Consent is also not transferable without limit. Permission for a film role does not necessarily cover political messaging, intimate content, product endorsements, or a voice assistant. Teams should request the narrowest rights needed and return for new permission when the purpose changes.

Identity can be conveyed through a face, voice, body, name, mannerisms, clothing, location, or a combination of clues. A generated character that does not exactly copy one photograph may still be recognizable as a particular person.

This is important for impersonation. A scam may use a cloned voice without video. A fake account may combine a real person’s name with generated photographs. A humiliating image may imply a person’s identity through context even if facial resemblance is imperfect.

Ask whether a reasonable audience could identify the person and attribute speech, conduct, or endorsement to them. Technical claims that “the pixels are new” do not settle the human question.

Fraud and operational harm occur when borrowed authority prompts payments, credential disclosure, account changes, or unsafe instructions.

Reputational harm occurs when media falsely associates someone with statements, behavior, or products. Corrections may not reach everyone who saw the original.

Dignitary and sexual harm occur when a person is placed into intimate or degrading material without consent. The harm includes loss of control, humiliation, harassment, and fear, whether or not every viewer believes the media is authentic.

Privacy harm occurs when source material was collected or reused beyond expectations, or when synthetic content reveals sensitive inferences and personal context.

Opportunity harm occurs when fabricated media affects employment, education, housing, relationships, or public participation.

Collective harm can affect communities through targeted stereotypes, intimidation, or confusion about public events. People with less access to legal, technical, or media support may carry greater recovery costs.

These harms can overlap. A fake voice message may create financial loss, workplace discipline, and emotional distress at the same time.

A prominent label can reduce audience confusion. It should appear before or during exposure, use plain language, and survive likely formats such as screenshots or clipped reposts where feasible. “Created with AI” may be too broad when viewers need to know that a real person’s voice was simulated.

Yet labeling does not authorize harmful use. A labeled fabricated endorsement still trades on someone’s identity. A labeled non-consensual intimate image remains non-consensual. A tiny footer does not cure a misleading headline.

Think of disclosure as one control alongside consent, purpose limits, access restrictions, review, complaint channels, and removal processes.

If someone reports impersonation or non-consensual synthetic media, begin with their safety and agency. Do not demand that they repeatedly view or publicly prove the material. Preserve only the evidence needed for reporting, according to safe procedures. Avoid forwarding the content except to authorized support, platform, legal, or law-enforcement channels when appropriate.

Document URLs, account names, dates, and relevant communications. Use platform reporting routes for impersonation, privacy, harassment, or non-consensual intimate imagery. In a workplace, involve the designated security, HR, legal, or safeguarding contact rather than improvising a public confrontation.

Do not blame the targeted person for having public photographs or recordings. The misuse is the responsibility of the person or organization that created, distributed, or exploited it.

Organizations can reduce risk before publication:

• use fictional identities when no real person is needed;
• collect purpose-specific permission in understandable language;
• restrict access to voice and likeness assets;
• prohibit sensitive or deceptive uses contractually and technically;
• require review before public release;
• make disclosure prominent and durable;
• provide a fast contact for correction, removal, and appeal;
• delete source material and derived assets according to a clear schedule.

The safest asset is often the one never collected. A team that needs a generic narrator does not need to clone a customer or employee.