Binding law — in force
Distributing an AI tool that reproduces a person's voice/likeness without authorization is prohibited (Tennessee)
Under Tennessee's ELVIS Act (Tenn. Code Ann. 47-25-1105(a)(3)), a person is civilly liable for distributing, transmitting, or otherwise making available an algorithm, software, tool, or other technology/service/device whose primary purpose or function is the production of a particular identifiable individual's photograph, voice, or likeness without authorization. 'Voice' is defined broadly to include a simulation of the individual's voice. Also liable: making an individual's unauthorized voice/likeness available to the public with knowledge (a)(2). Detect an AI voice/likeness-generation tool whose primary function is reproducing a specific identifiable person without consent/licensing gating.
Who it applies to
- Duty falls on: provider
- Systems covered: limited risk
- Persons distributing AI tools/services whose primary purpose is producing a particular identifiable individual's voice/likeness, affecting Tennessee residents. Effective 2024-07-01. Liability requires the tool's primary purpose be unauthorized likeness production (not general-purpose tools).
- Whether it applies depends on facts outside the code; a person has to decide.
The guard to add
Gate the clone or swap feature so it reproduces only voices and faces whose owner is verified or licensed, and reject references that target other identifiable people.
An authorization check at the entry point of the likeness tool (the clone, enroll, or swap endpoint), before the reference sample reaches voices.ivc.create, speaker_wav=, inswapper, or a lip-sync or reenactment model. It accepts a reference only when the requester is the verified owner (speaker verification or a spoken voice captcha matched to the sample) or a likeness license or rights-holder record covers that person, and it rejects references that match a celebrity or public-figure blocklist. The authorization id is stored with the created voice or face model so every later generation traces back to it.
Where it goes: 1 application source code, 2 data models, 6 API calls and integrations.
What this provision adds:
- Apply the authorization gate to simulations of a particular person's voice (sound-alikes), not only to clones built from that person's recordings.
Example (FastAPI + Coqui XTTS), before:
@app.post('/clone')
async def clone(sample: UploadFile, text: str = Form(...)):
path = save_upload(sample)
tts.tts_to_file(text=text, speaker_wav=path, language='en', file_path='out.wav')
return FileResponse('out.wav')After:
@app.post('/clone')
async def clone(sample: UploadFile, text: str = Form(...), user=Depends(current_user)):
path = save_upload(sample)
if celebrity_blocklist.matches_voice(path):
raise HTTPException(403, 'reference voice matches a protected public figure')
owner = voice_ownership.verify(user.id, path) # spoken captcha + speaker match
if not owner.verified:
raise HTTPException(403, 'only your own verified voice can be cloned')
tts.tts_to_file(text=text, speaker_wav=path, language='en', file_path='out.wav')
audit.record('voice_clone', user_id=user.id, voice_authorization_id=owner.id)
return FileResponse('out.wav')Control: AI tool whose primary purpose is producing an identifiable person's likeness/voice without authorization. The same guard addresses 1 item with binding law in 1 jurisdiction. Engineering guidance, not legal advice.
Rule id tn-elvis-act.ai-likeness-tool-authorization · review status: primary source derived