🔍 Read the full analysis: Human-Sounding AI Is Here. What Block’s Dessa’s Voice Can Tell Us on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Block announced Dessa, a speech-to-speech research model, on October 23, 2026, and says it was trained from scratch on recorded human conversation. The demos show expressive, interactive speech, but no independent testing, public model access, technical paper or benchmark results are available. The model’s potential naturalness also sharpens concerns about voice impersonation and fraud.
Block, the company behind Square and Cash App, announced Dessa, a research model that generates speech while listening and responding in real time. Its selected demos feature laughter, interruptions and emotional-sounding replies, but the model has not been independently tested and is not publicly available, leaving its capabilities and safeguards unverified.
Block says Dessa was trained from scratch on recorded human conversation, without text or pretrained models. The company describes it as a full-duplex system: it can listen while speaking, respond to pauses and interruptions, and produce cues such as laughter or a frustrated sigh. Block also says inference runs faster than real time on a MacBook Pro M4 Max or better, but has not published a latency figure.
The public evidence consists of recordings selected and published by Block. The examples include a model laughing mid-sentence, helping a speaker remember a movie title and responding to news of an anniversary surprise. Block says each clip is a live conversation and that nobody steered the model. That description has not been independently verified.
Outside assessments have been more cautious than the launch language. The Verge reported on October 23 that some exchanges were awkward or uncanny, while describing Dessa as more natural than another model in one comparison. TechCrunch noted the uncertainty over whether emotion might be steered by tags. Block has published no technical paper, model card, benchmarks, weights or API; access is limited to a waitlist and a small group of AI safety researchers.
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 Speech and Voice Fraud
A conversational voice that handles pauses and emotional cues could make voice interfaces feel less mechanical. The same qualities may make it harder for a listener to tell whether they are speaking with a person or a synthetic voice. That risk matters because voice impersonation scams already exist; Dessa is not needed for them to occur.
The source material cites the FBI’s 2025 Internet Crime Report, published 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 cloned voices of loved ones. These figures describe reported complaints and losses, not the performance or use of Dessa, and do not establish that the model has been used in fraud.
For organizations, the development is a reminder that a familiar voice alone is weak proof of identity. Payment requests should be verified through a separate, known channel. For families, a shared code word and calling back on a saved number can help check urgent requests. These are practical precautions, not evidence that Dessa itself has been misused.
Products Worth Considering
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
What Block Has Disclosed
Block published the project page on October 23, 2026. Co-founder Jack Dorsey described talking with the model as feeling like talking with a person, citing emotion, pauses and natural cadence. Dessa lead Ragavan Thurairatnam called it “a GPT3 moment for speech” and said it could run on consumer GPUs. Those are attributed assessments, not independent benchmark results.
Block says the model was trained using a small budget and one NVIDIA GB200 NVL72, and that it learned from hundreds of times fewer words than large language models require. The company has not provided enough technical detail to independently evaluate those comparisons. The team is described as a small Block research lab formed by former members of Toronto-based Dessa, which Square acquired in 2020.
The model’s name and expressive demos have prompted opposing reactions. Some listeners praised its realism; filmmaker Justine Bateman questioned why AI voices should sound human, arguing that this could lower people’s guard. Block frames emotional speech as a way for voice systems to better respond to users. The competing interpretations concern the same capability, but neither establishes how the model will perform in real-world products.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
— Jack Dorsey, Block co-founder
Key Tests and Safeguards Missing
Dessa’s performance remains unverified: no outside researcher has run the model, and no public weights or API are available. Block has not disclosed its size, benchmark results, a public release date or a technical paper. Its claim that the model runs faster than real time also lacks a published latency measurement.
It is also unclear which languages it supports, whether its emotional delivery relies on steering tags, or whether it can imitate a particular person. The source material reports no disclosed policy or technical details on watermarking, consent checks or safeguards against impersonation. A natural-sounding demo does not establish that the system has feelings, general intelligence or dependable performance in ordinary conversations.
The FBI figures describe reported losses and complaints, not a measured increase caused by Dessa or a forecast of future fraud. The available material does not show that the model has been used in scams. Any connection between the research announcement and specific criminal activity would be speculation.
Waiting for Independent Access
The next useful evidence would be independent access, technical documentation and reproducible tests. Block has offered a waitlist and early access to a small group of AI safety researchers, but has not announced when broader access or a public release might happen. Until then, outside assessments are confined to the company’s recordings.
Researchers and prospective users will need details on languages, latency, voice imitation, consent and disclosure controls to judge how the model works and where it can be used safely. Businesses should continue verifying sensitive instructions through separate channels rather than relying on a caller’s voice. Families can agree on a code word and call back using a number they already trust, regardless of whether a particular AI model is involved.
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, responding to conversational cues such as pauses and interruptions.
Can the public use Dessa now?
No public model, weights or API have been announced. Block has a waitlist and limited early access for a small group of AI safety researchers.
Has anyone independently verified the demos?
Not by running the model. The available examples were published by Block, and no outside testing of the model is reported in the source material.
Does Dessa prove that voice AI has feelings?
No. An emotional-sounding voice is not evidence that a model experiences emotion. The published demonstrations show speech behavior, not feelings or human understanding.
What can people do to reduce voice-scam risk?
For urgent requests involving money or sensitive information, hang up and call back using a known number. Families can also agree on a code word. These steps help verify identity without relying on a voice alone.
Source: ThorstenMeyerAI.com
