Docs and sources: OpenAI didn't use Scarlett Johansson's voice for ChatGPT's Sky and hired another actress in June, months before Sam Altman contacted Johansson
A different actress was hired to provide the voice for ChatGPT's ‘Sky,’ according to documents and recordings shared with the Washington Post.
Washington PostNitasha Tiku
Context & Ripple Effects
The dispute escalated after OpenAI withdrew Sky amid comparisons to Johansson and Johansson said she had declined an offer to voice the product. The Washington Post’s reported documents and recordings add a provenance-based account to what had largely been a public similarity dispute.
The controversy arrived as OpenAI was demonstrating a more conversational assistant that could handle interruptions and expressive exchanges. That makes the origin and selection of voices consequential to trust in its end-to-end speech experience, not merely a branding detail.
First-order effects
The reported hiring records and recordings bolster OpenAI’s position that Sky was performed by a different actress rather than built from Johansson’s voice.
The evidence shifts immediate scrutiny toward the selection process, internal communications, and why the released voice was perceived as so similar, even if the underlying performer was different.
Second-order effects
Voice-AI developers face stronger pressure to document performer consent, provenance, and approval trails before launch, particularly for voices likely to evoke recognizable public figures.
Product teams may treat perceived vocal resemblance as a release risk independent of whether a model or actor literally copied a celebrity, increasing review burden for consumer-facing voice options.
The episode points to a broader boundary problem for synthetic media: consent-based sourcing may not by itself resolve disputes when a deployed output creates a strong public association with a specific person.
The trend: Conversational AI is moving toward more humanlike voice interfaces, making identity, consent, provenance, and perceptual similarity central product-governance issues.
Wow... that really sounds like a deliberate ploy on OpenAI's part to embarrass Johansson. — And frankly, sounds like exactly the shitty kind of thing that Silicon Valley techbro incels would take particular glee in. [embedded post]
OpenAI didn't copy Scarlett Johansson's voice for ChatGPT, records show A lot of people who should have known better rushing to judgement here without the facts... https://www.washingtonpost.com/ ...
Paul Graham wrote an essay in 2010 about how the best founders are naughty because they break the rules Then he has an interesting anecdote about Sam Altman... Interesting lens through which to view why OpenAI keeps showing up in the news for doing obviously bad things [image]
The lede of this story seems to exonerate OpenAI but buried in the piece is how the company is far, far from off the hook, assuming this mystery actress is legit. https://www.washingtonpost.com/ ...
new from me & @pranshuverma_ talked to the casting directors, Sky's voice agent and listened to some tape to figure out how OpenAI made the AI voice that sounds like ScarJo if you read the thinkpieces stop thunking already and read our reporting (pls) https://www.washingtonpost.c…
@yoda @harrymccracken Right, I still don't necessarily buy that they didn't try to mimic her voice. And ScarJo may still have a case. But the order of events is different than people assumed, rendering a lot of the more heated takes inaccurate imo.
All this tells me is that these ChatGPT dudes are such droids that they don't think likeness-infringement happened as long as it wasn't a computer that did it. Try again, nerds. Extensive body of case law that says quite the opposite https://www.washingtonpost.com/ ...