🔍 Read the full analysis: What Block’s Dessa Reveals About AI That Sounds Human 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 research model that generates and responds to speech, and published recordings designed to show conversational features such as laughter and interruptions. The model has not been independently tested or publicly released, so its capabilities remain company claims; its human-like voice also renews concerns about impersonation fraud.
Block announced Dessa, a speech-to-speech research model, on October 23, 2026, and published recordings in which it laughs, responds to interruptions and speaks with apparent emotion. The demonstrations present the company’s approach to conversational voice AI, but no outside researcher has tested the model, and Block has not released technical documentation, model weights or an API.
Block says Dessa was trained from scratch on recorded human conversation alone, without text or pretrained models. The company describes it as a full-duplex system: it can listen while speaking, interrupt or respond with short conversational signals, and process qualities such as laughter, sarcasm and changes in tone. These descriptions come from Block; the company has not published a paper or benchmark results to support an independent assessment.
The project page includes company-selected recordings of conversations, including a moment where the model helps a user recall a film title and another in which it responds to news of an anniversary surprise. Block says the clips are live recordings and that nobody steered the conversations. The Verge described some exchanges as occasionally awkward and uncanny, while finding the system more natural-sounding than another model in one comparison. TechCrunch said it was unclear whether emotional delivery might be guided by tags.
Block says inference runs faster than real time on a MacBook Pro M4 Max or better, but has not provided a latency figure or disclosed the model’s size. The company says training used one NVIDIA GB200 NVL72. Access is limited to a waitlist and a small group of AI safety researchers; Block has not announced a public release date, license or API.
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
Human-Sounding Voices and Fraud Risk
Dessa’s demonstrations show conversational features such as pauses, laughter and responses to emotional exchanges. These qualities may make automated speech less readily distinguishable from human speech, although the recordings do not establish that Dessa can imitate a particular person or has been used in fraud. The demonstrations raise questions also associated with voice-cloning tools more broadly.
The FBI’s 2025 Internet Crime Report, published in April 2026, counted 22,364 complaints mentioning AI and $893 million in reported losses. It also reported more than $5 million in losses from distress scams involving cloned voices of loved ones. Those figures cover reported crimes, not all incidents, and do not attribute losses to Dessa. They illustrate why a familiar-sounding voice alone should not be treated as proof of identity.
Businesses and families can use verification steps that do not depend on how a caller sounds. Companies can require callbacks to known numbers for payment requests and establish approval procedures for financial changes. Individuals can agree on a family code word and end a suspicious call before calling back using a number they already trust. Urgency and requests for secrecy can be warning signs, even when a voice sounds familiar.
Popular right now
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
TCL 85 Designer Series QD-Mini LED TV (Model 85A400U) with Google Smart TV
Breville Bambino Espresso Machine, Brushed Stainless Steel
As an affiliate, we earn on qualifying purchases.
What Block Has and Has Not Shown
Block, the company behind Square and Cash App, published the Dessa project page on October 23. Co-founder Jack Dorsey described speaking with the system as feeling like talking with a person, citing emotion, nuance, pauses and natural cadence. Dessa lead Ragavan Thurairatnam called the development “a GPT-3 moment for speech” and said it could run on consumer GPUs. These are statements from Block’s leadership, not independent findings.
The project comes from a small research lab at Block founded by former members of Dessa, a Toronto deep-learning startup acquired by Square in 2020, according to CoinDesk. Block presents the system as an alternative to voice systems built around converting speech to text and then generating a response. The company says those pipelines lose emotional information. Whether Dessa preserves such information reliably across varied speakers, settings and languages has not been established in public tests.
As of October 26, the public evidence consists of Block’s recordings and limited media reviews. No technical paper, model card, benchmark suite or public model access was available. Block has also described learning new languages efficiently as an open question. The demonstrations therefore show what the company chose to present, not a verified account of performance across ordinary calls or high-stakes situations.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
— Jack Dorsey, Block co-founder
Independent Tests Are Still Missing
It remains unclear how well Dessa performs outside the recordings Block released. There is no public benchmark data, technical paper or model card, and outsiders have not been able to run the model. The model’s size, language coverage, failure rates and behavior in noisy or emotionally difficult conversations have not been disclosed.
Block says Dessa selects its own voice and personality, but has not publicly explained whether it can imitate a specific person, what safeguards prevent that, or whether its output is watermarked. The company has not described consent checks or detailed access controls. It is also unclear when or whether the model will become broadly available. The available evidence does not establish that the model is conscious, understands emotion as a person does, or has been used in a scam.
Release Details and Safeguards to Watch
Independent researcher access, accompanied by documentation on training, evaluation, limitations and safeguards, would allow further assessment of Block’s claims. Block says a small group of AI safety researchers has early access, but has not shared their findings or set a date for a wider release. Public access would also allow reproducible tests of how the model handles interruptions, emotional cues and different languages.
Until further information is available, the demonstrations remain a company presentation rather than independently verified evidence of general reliability. Businesses evaluating voice AI can set identity checks that do not depend on voice alone, while families can establish a verification phrase and a trusted callback routine. The public record may change as Block publishes more information or outside researchers report their findings.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says can listen and speak at the same time. The company says it was trained from scratch on recorded human conversation.
Can the public use or test Dessa now?
No public release has been announced. Block says access is limited to a waitlist and a small group of AI safety researchers. The company has not announced public model weights or an API.
Has an independent group confirmed Block’s claims?
Not as of October 26, 2026. The available demonstrations were published by Block, and no public paper, model card or benchmark results were available. Media outlets have listened to the clips, but have not independently run the model.
Does Dessa clone specific people’s voices?
That has not been established. Block says the model chooses its own voice and personality, but has not publicly explained whether it can imitate a specific person or what safeguards address that possibility.
How can people check a suspicious voice call?
Do not rely on the caller’s voice alone. Hang up and call a number you already know, and for families, agree on a private code word. For payment requests, businesses can use a second approval channel and verify instructions through established contacts.
Source: ThorstenMeyerAI.com
