Sunday, 1 November 2026 Our record: 59% called, 80 calls on file SearchSubscribe
Crypto and finance news. On the record.
Breaking
AI3 min read

Block Dessa voice AI wows testers and fuels phone scam fears

Jack Dorsey's payments company says its new model learned to talk only from human speech and runs on a laptop. Listeners were impressed, and fraud watchers were alarmed.

Block Dessa voice AI wows testers and fuels phone scam fears
Illustration: Called It

The Block Dessa voice AI has drawn both awe and alarm since its debut. Jack Dorsey's payments company unveiled Dessa on Friday, Oct. 23, describing it as a speech-to-speech model trained entirely on human conversation rather than text, according to CoinDesk and The Verge.

What happened

Most voice assistants convert speech to text, generate a written reply and then read it aloud. Dessa works on audio directly and can listen while it talks, so it handles interruptions, laughter and changes of tone mid-sentence, Block said. The company trained it from scratch on a single Nvidia GB200 NVL72 system, and the model is small enough to run on a MacBook Pro fitted with at least an M4 Max chip, CoinDesk reported.

Block called Dessa "the world's first AI you can have a realistic live conversation with that was trained only from human speech," PYMNTS reported. In one demo the model laughs in the middle of a sentence, and in another it responds to a surprise with what sounds like real emotion; Block says the conversations were recorded live, with no one steering the replies. Other samples on the Dessa website include a chat about Christopher Nolan's films that wanders into Bryan Singer's Superman movies, and a user announcing a surprise trip to Paris, which the site says Dessa greets with "teary gratitude," The Verge reported. Dorsey wrote on X that "Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence."

The demos are not flawless. In one clip shared by Dessa lead Ragavan Thurairatnam, the model mistakes the speaker for someone named "Ronaldo," The Verge noted. Another pits Dessa against OpenAI's GPT-Live-1 voice model, with Block's system coming across as the more natural of the two.

Why it matters

The reaction split quickly. Binance founder Changpeng Zhao said the model could suit children's education: "This might be very good for GiggleAcademy," he wrote, according to PYMNTS. Filmmaker Justine Bateman warned that a human-sounding voice could "trick people into lowering their guard," and journalist Shiv Aroor said scam calls were "about to attain pandemic level."

Those fears are concrete for a payments firm. Voice cloning and impersonation already drive fraud that regulators are trying to contain, a theme in our report on the White House crypto scam crackdown. A model that runs on a laptop, without a cloud provider able to monitor usage, would be harder to police once released. Block's decision to limit the first research preview to a small group of AI safety researchers suggests it knows this.

The Block Dessa voice AI also shows how quickly capable models are shrinking. Big labs are racing to ship ever-larger frontier systems, such as Meta's Muse models, while Block argues that a focused model trained on the right data can feel more human on far less hardware.

What's next

PYMNTS has argued that voice assistants have largely solved understanding what people say, and that the contest now is whether they can do something useful with it. Investors were unmoved for now: Block shares closed at $76.45 on Friday, down 3.25%, CoinDesk's market data showed.

Block has not disclosed the model's size, benchmarks or a public release date, CoinDesk said. Those details, plus any safeguards against cloning real voices, will decide whether the Block Dessa voice AI becomes a product or stays a research showcase. The reactions are collected in CoinDesk's report.

This article is for information only and is not investment advice.

More from AI

All ai

The Morning Call.

The day's crypto and finance news, one call and one chart. Weekdays at 7am ET.