Statement from Scarlett Johansson on the OpenAI "Sky" voice
OpenAI’s “Sky” voice for ChatGPT, which many felt closely resembled Scarlett Johansson’s AI character in the film *Her*, has triggered a backlash after Johansson revealed the company twice sought — and failed to obtain — her permission before launch. Commenters debate whether this could violate U.S. “right of publicity” and voice-impersonation case law, drawing parallels to past lawsuits over celebrity soundalikes. Beyond legal risk, many see the incident as emblematic of OpenAI’s and its CEO’s cavalier approach to consent, artists’ rights, and public trust in companies building powerful AI systems.
Alleged voice cloning & timeline
- Discussion centers on the “Sky” ChatGPT voice sounding like Scarlett Johansson’s AI character in Her.
- Commenters recap the timeline in Johansson’s statement:
- OpenAI approached her last fall; she declined.
- Her team was asked again two days before the GPT‑4o demo.
- OpenAI launched anyway, CEO tweeted “her”, and only removed the voice after her lawyers contacted them.
- Many see this as strong evidence of intent to evoke her likeness; some say the CEO’s tweet “gave away the game”.
- A minority say they never heard the resemblance and feel social media hype created the association post‑hoc.
Legal framing: likeness and “soundalikes”
- Multiple commenters cite US “right of publicity” / personality rights, particularly California law, as the likely basis for a claim.
- Past cases frequently mentioned: Bette Midler vs Ford, Tom Waits vs Frito‑Lay, Vanna White vs Samsung, plus other voice/likeness cases.
- Key idea: you can’t commercially exploit a famous, distinctive voice via an impersonator after being refused, even if you never say it’s that person.
- Others worry about edge cases: what if someone naturally sounds similar, or parody / satire; some fear overbroad rights could harm unknown actors.
Ethics, consent, and OpenAI’s conduct
- Strong sentiment that asking twice, being refused, then deploying a sound‑alike (and pulling it only under legal pressure) is ethically “beyond the pale”.
- Many connect this to broader patterns: training on unlicensed copyrighted data, opaque datasets, aggressive NDAs, and the disbanding of safety teams.
- Several say this undermines OpenAI’s self‑presentation as a responsible AI “safety” leader and will erode public and legislative trust.
Technical / product angles
- “Sky” has existed for months; GPT‑4o mainly made it more expressive (laughter, singing, flirtiness), which intensified the Her association.
- Some argue they could have used any generic voice; the added legal/PR risk for negligible product benefit seems irrational.
Broader implications & divided views
- Many hope for a lawsuit to set precedent on AI voice/likeness use and to open discovery into OpenAI’s training data.
- Others think OpenAI will quietly settle; a few are cynical that fines will just be “cost of doing business”.
- A smaller group defends OpenAI, arguing that similarity alone shouldn’t be illegal and that companies often use stylistic soundalikes.