Meta gave its assistant five actors' licensed voices, not clones

What happened

On 25 September 2024, Meta published a Connect keynote recap stating that Meta AI would gain a voice, and that 'as this feature starts to roll out,' users would 'choose from different voice options for your assistant, including some familiar ones like the AI voices of Awkwafina, Dame Judi Dench, John Cena, Keegan Michael Key and Kristen Bell.' The post does not give a rollout date beyond 'starts to roll out,' nor does it name a contract length or the countries covered.

What the documents show

The announcement's own phrasing, 'the AI voices of' five named performers, describes a licensed voice: a company obtaining rights from an identified, credited individual to use a synthesized version of their voice, distinct from a voice cloned from an anonymous or unconsenting source. Checking Meta's own current product page, ai.meta.com/meta-ai, as retrieved on 16 September 2026, finds no mention of these named voices; the page instead describes Meta AI in general terms as a 'personal AI assistant.' That silence does not establish that the feature was withdrawn, changed, or renamed, only that the specific voice options named in 2024 are not described on the current page reviewed.

The mechanism

A licensed celebrity voice sits at one end of a spectrum of ways a synthetic voice can relate to a real person, alongside a self-service clone built from someone's own uploaded recording and a fully synthetic voice tied to nobody in particular. What distinguishes Meta's approach is that the relationship is individually negotiated and publicly credited: the announcement names each performer and attributes the voice to them by name, which is a different kind of claim than a platform's general terms permitting any user to clone whichever voice they are willing to attest they may use. The credited, contracted structure is the mechanism worth naming when a company markets a 'celebrity voice' rather than treating all synthetic voices as interchangeable.

What it leaves open

Neither document establishes how long the arrangement with these five performers ran, what markets received it, or what became of the feature after its 2024 rollout. This is an evidentiary gap the record should flag rather than fill: the 2026 product page's silence is consistent with several different explanations that the sources opened here do not settle.

• Is a marketed 'celebrity voice' individually licensed and credited, or generated by a general-purpose cloning tool?
• What does a company's silence on a previously announced feature, in its own current documentation, actually establish?
• Who bears responsibility if a licensed voice is later used in a context the performer did not approve?

The 2024 announcement is solid evidence of what Meta said it would ship; it is not evidence of what remained in place two years later, and the record should hold that distinction rather than assume continuity.