🔍 Read the full analysis: AI That Laughs And Hesitates: Block’s Dessa Puts Voice In Perspective on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Block announced Dessa, a speech-to-speech research model trained on recorded human conversation, and released selected demos showing laughter, pauses and interruptions. The model has not been made publicly available or independently tested, so its capabilities and safeguards remain unverified; its human-like voice also sharpens concerns about impersonation scams.
Block announced Dessa, a speech-to-speech research model, on October 23, saying it was trained from scratch on recorded human conversation and can listen while speaking. Selected demos show it laughing, pausing and responding to interruptions, but no independent researcher has tested the model, and Block has not released its weights, a paper or benchmark results. The demonstrations matter because natural-sounding voices could make conversational AI more useful—and make voice impersonation harder to spot.
Block, the company behind Square and Cash App, describes Dessa as a model trained on audio rather than text, without pretrained models. The company says it can recognize cues such as sarcasm, laughter and a frustrated sigh, and handle overlapping turns by listening while it talks. It also says inference runs faster than real time on a MacBook Pro M4 Max or better, but did not provide a latency measurement.
The available evidence consists of recordings selected and published by Block. In them, Dessa laughs mid-sentence, helps a speaker recall a movie title and responds to an anniversary surprise. Block says each clip is a live recording and that nobody steered the model. The company has not published a technical paper, model card, benchmark results, license or release date. There is no public API or downloadable model; Block says a small group of AI safety researchers can access it, while others can join a waitlist.
Outside accounts of the demos were more qualified than the launch description. The Verge reported that the exchanges could be awkward and uncanny, while also finding Dessa more natural-sounding than another model featured in the material. TechCrunch said it was unclear whether emotion in the output might be steered by tags. Those observations concern the recordings, not tests of the model itself, which remains unavailable for independent evaluation.
Infographic 01 · Dessa fact sheet
What Block claims, what was heard, what is missing.
Every capability below comes from Block or its selected recordings. Nobody outside Block has run the model yet.
| Topic | What we know | Status |
|---|---|---|
| Training data | Human conversation only: “no text, no pretrained models” | Block’s claim |
| Conversation | Full-duplex: listens while it talks, backchannels, cuts in | Block’s claim |
| Emotion | Hears and speaks laughter, sarcasm, sighs; unclear if steered by tags (TechCrunch) | Block’s claim |
| Speed | “Faster than real time”; no latency figure | Block’s claim |
| Hardware | Trained on one NVIDIA GB200 NVL72; runs on a MacBook Pro M4 Max or better | Block’s claim |
| Demos | Some chats “a little awkward and uncanny”, more natural than GPT-Live-1 (The Verge) | reporters heard it |
| Size, benchmarks, release date | Not disclosed (CoinDesk) | not disclosed |
| Paper, model card, license, weights | None found as of 26 Oct 2026 | not disclosed |
| Languages | Not listed; Block calls learning new languages an open question | not disclosed |
| Voice cloning safeguards | No statement on watermarking, consent checks or imitating a specific person | not disclosed |
Who can use it
research preview
- Public: a waitlist; “launch links are coming soon”
- Early access: “a small group of AI safety researchers”, institutional email required
- Business use: none announced; Square already uses voice AI for ordering (TechCrunch)
Who built it
from Block’s page
- A “(very) small AI research lab at Block” reporting to Jack Dorsey
- Founded by former members of Dessa, the Toronto startup Square acquired in 2020 (CoinDesk)
- Lead Ragavan Thurairatnam: “a GPT3 moment for speech”
Sources: Block, Dessa project page (23 Oct 2026); Ragavan Thurairatnam on X (23 Oct 2026); The Verge (23 Oct 2026); TechCrunch (24 Oct 2026); CoinDesk (24 Oct 2026).
Powered by Thorsten Meyer AI
Natural Voices Raise Fraud Risks
Dessa’s demonstrations highlight a tension between conversational quality and voice-based deception. Laughter, pauses and responsive backchannels may make voice assistants feel more natural. The same cues could make an impersonation more convincing if a person assumes a familiar-sounding caller is genuine. The demos do not establish that Dessa can copy a specific person, and Block has not said whether the model includes safeguards against imitation.
The source article cites the FBI’s 2025 Internet Crime Report, released in April 2026, which counted 22,364 complaints mentioning AI and $893 million in reported losses. It also says victims lost more than $5 million to distress scams involving voice cloning of loved ones, while businesses lost more than $30 million to AI-assisted email-compromise scams. These figures cover reported crimes in 2025; they do not measure losses caused by Dessa or show that its demonstrations have been used in fraud.
For businesses, the development is a reminder that a voice alone is not reliable proof of identity when money, account access or sensitive information is at stake. The source advises verifying payment requests through a known callback number and adopting disclosure practices for synthetic voices. For families, an agreed code word and a call back to a saved number can add checks when a caller claims an emergency. Those are precautionary measures, not evidence that Dessa itself has been misused.
Popular right now
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
Samsung 48-Inch S90H OLED 4K Smart TV (2026)
As an affiliate, we earn on qualifying purchases.
From Dessa Startup to Block Research
The name links Block’s project to Dessa, a Toronto deep-learning startup that Square acquired in 2020, according to CoinDesk. Block describes the current effort as a small research lab founded by former members of that startup. The project page was published October 23, 2026; co-founder Jack Dorsey and research lead Ragavan Thurairatnam described the model in public statements, but the company has not released a full technical account.
Block’s pitch centers on preserving information that ordinary speech-to-text systems may discard, such as timing and vocal emotion. The company also says Dessa learned language from far fewer words than large language models typically use, but it has not disclosed the model’s size or supplied comparisons that would let outsiders assess that claim. How well it handles different languages is also an open question, according to the source material.
The gap between a polished demo and a tested system is central here. Reporters have listened to Block’s recordings, but the model is not available for outside testing. The recordings can show how selected conversations sound; they cannot establish performance across users, accents, languages or difficult conditions. Nor do they establish that the system is emotionally aware: sounding emotional is not proof of feeling or understanding.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
— Jack Dorsey, Block co-founder
Testing and Safeguards Remain Open
As of October 26, no outsider has run Dessa, and Block has not released the technical details needed to evaluate its claims. The model’s size, benchmark performance, supported languages, public availability and release timing are unknown. It is also unclear how it performs beyond the selected recordings, including when speakers interrupt, use unfamiliar accents or speak in noisy settings.
Block says the model chooses its own voice and personality, but has not publicly detailed whether it can imitate a particular person or what controls prevent misuse. The source material also found no public statement on watermarking or consent checks. Without access to the model and information about its safeguards, the demos cannot establish how easily it could be used for impersonation—or how effectively such use could be detected.
The source article says Article 50 of the EU AI Act applies from August 2, 2026, and identifies disclosure of synthetic content as relevant to companies deploying voice AI. That does not resolve the specific obligations for every use of Dessa: Block has not announced a deployment, and the project is not publicly available. Its eventual product form and compliance measures remain unknown.
A Release Would Enable Testing
The next meaningful milestone would be the release of a paper, model card, benchmarks or wider research access. Those materials could help independent researchers check Block’s claims about speed, conversation handling, language performance and safeguards. Block has not announced when any of them will appear, and it has not provided a public release date for the model.
Until then, the available record is limited to Block’s project description and selected audio demos, along with reporters’ observations of those recordings. Companies considering voice AI will need to wait for clearer information on consent, disclosure and identity protections before drawing conclusions about deployment. People receiving urgent requests by phone can use existing verification habits—such as calling back a known number—without assuming that a natural-sounding voice proves who is speaking.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says was trained from scratch on recorded human conversation. The company says it can listen while speaking and respond to vocal cues such as laughter and sarcasm.
Can the public use or download the model?
No public weights or API have been announced as of October 26, 2026. Block says a small group of AI safety researchers has early access and offers a waitlist, but it has not given a public release date.
Has Dessa been independently tested?
No outsider has run the model, according to the source material. Reporters have listened to Block’s selected recordings, but there is no public benchmark or technical paper for independent evaluation.
Has Block shown that Dessa can clone a specific person?
That has not been established. Block says the model picks its own voice and personality, but has not publicly explained whether it can imitate a particular person or detailed safeguards against imitation.
What precautions can people take against voice scams?
For urgent requests involving money or personal information, hang up and call back using a number already known to you. Families can also agree on a code word. These are general precautions; the source does not report that Dessa has been used in a scam.
Source: ThorstenMeyerAI.com
