🔍 Read the full analysis: When A Voice Sounds Human, What Counts As Proof? Block’s Dessa on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Block announced Dessa, a speech-to-speech research model, on October 23, 2026, and says it can listen and respond in real time with laughter, emotion and interruptions. The public evidence remains limited to Block-selected demonstrations: there is no published paper, benchmark, model card or general public access. Its humanlike voice renews questions about voice-based fraud and how businesses should verify callers.
Block announced Dessa, a speech-to-speech research model, on October 23, showing recordings in which it laughs, responds to emotion and talks over a speaker naturally. The company says the model was trained from scratch on recorded human conversation and can run faster than real time on certain high-end Mac computers, but no independent testing or public release has been reported, limiting what outsiders can verify about its capabilities.
Block, the company behind Square and Cash App, describes Dessa as a model that listens and speaks at the same time, rather than converting speech to text and then generating a spoken response. Its project page says it was trained on human conversation without text or pretrained models, and that it can recognize and produce features such as laughter, sarcasm and pauses. Those are company claims; Block has not published a technical paper or benchmark results to substantiate them.
The company’s recordings include laughter during a sentence, a conversation about a forgotten film title and a response to an anniversary surprise. Block says the clips are live conversations and were not steered. The Verge, after listening to the demonstrations, described some moments as “a little awkward and uncanny”, while saying Dessa came across as more natural than another model featured in the comparison. TechCrunch reported that 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, and that training used one NVIDIA GB200 NVL72. It has not provided a latency figure, disclosed the model’s size or announced a release date. Access is limited to a waitlist and a small group of AI safety researchers, according to the source report; weights and an API have not been announced.
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
Humanlike Speech Raises Verification Risks
Dessa’s demonstrations matter beyond whether the model sounds natural: expressive, responsive speech could make it harder for listeners to judge whether they are talking to a person. The source report cites the FBI’s 2025 Internet Crime Report, published in April 2026, which recorded 22,364 complaints mentioning AI and $893 million in reported losses. The FBI also reported more than $5 million in losses from distress scams involving cloned voices of loved ones. These figures describe reported cases and losses, not harms attributed to Dessa.
Voice imitation fraud predates this model. The concern raised by more natural conversational systems is that timing, interruptions and emotional cues may remove some clues people rely on when a call sounds artificial. That is a risk assessment, not evidence that Dessa has been used in a scam. The report says Block has not clarified whether the model can imitate a specific person or what safeguards it may have.
For businesses, the development is a reminder that a familiar-sounding voice should not, by itself, authorize payments or account changes. Verification through a separate, known channel can reduce reliance on caller identity. For families, agreeing on a private check phrase and calling back using a saved number are practical precautions against urgent impersonation attempts. Neither measure depends on proving that a particular call was generated by AI.
Products Worth Considering

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black

Bose QuietComfort Wireless Noise Cancelling Headphones – Twilight Blue Limited Edition

Apple MacBook Air 15-inch (2026) with M5 chip, 16GB memory, 512GB SSD — Midnight
As an affiliate, we earn on qualifying purchases.
What Block Has—and Has Not—Shown
Block’s project page was published on October 23, 2026. Co-founder Jack Dorsey praised the conversational quality, writing that talking with Dessa feels like talking with a person, with emotion, nuance, pauses and natural cadence. Dessa lead Ragavan Thurairatnam called the work “a GPT-3 moment for speech” and said it could run on consumer GPUs. These are statements from people involved in the project, not independent evaluations.
Outside reviewers have heard the company’s recordings but have not been able to test the model themselves. The source report found no technical paper, model card, benchmark, license or public release date as of October 26. It also found no published list of supported languages and said Block treats how efficiently Dessa can learn new languages as an open question. A demonstration can show what happened in selected clips; it cannot establish performance across users, situations or languages.
The distinction is particularly relevant because a convincing voice is not evidence of a speaker’s identity. The same source report notes that the FBI documented AI-related complaints and losses in 2025. That broader fraud data establishes a real risk environment, but it does not show that Dessa contributed to those incidents or that its specific capabilities are already available to the public.
“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 and Safeguards Are Missing
Key details remain unavailable as of October 26, 2026. Block has not released Dessa for general use, published weights or an API, or supplied independent benchmark results. Its training data, model size, supported languages and measurable response latency have not been fully disclosed in the source material. Without access, outside researchers cannot reproduce the demonstrations or assess how consistently the model performs.
It is also unclear whether emotional delivery is generated entirely by the model or partly guided by controls such as tags. Block says Dessa chooses its own voice and personality, but the report found no clear statement about watermarking, consent checks or person-specific voice imitation. Those unanswered questions do not establish that the model can clone a particular voice or that it lacks safeguards; they mean the public record does not yet show how those issues are handled.
Finally, recordings cannot establish how the system behaves in high-stakes calls, across accents or languages, or outside the selected scenarios. Block’s claims about speed and training scale also lack published measurements in the material available. Until testing and documentation are released, conclusions about reliability, safety and real-world risk should remain provisional.
Research Access and Documentation
The next meaningful evidence would come from Block publishing technical documentation and allowing independent researchers to test the model. The company has said a small group of AI safety researchers has early access, but the source report does not give a timetable for wider access, a paper or a product release. No subsequent milestone is confirmed.
Researchers and users will need details on evaluation methods, supported languages, voice consent and controls that could limit impersonation. Until those are available, Block’s recordings establish what the company chose to demonstrate—not how Dessa performs broadly or what protections accompany it. In the meantime, callers and organizations can treat voice alone as insufficient proof of identity and verify urgent requests through a separate, previously trusted channel.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model announced by Block on October 23, 2026. Block says it listens and responds directly in speech and can handle conversational cues such as laughter and interruptions.
Has Dessa been independently tested or released publicly?
Not according to the source material available as of October 26, 2026. Block has shown recordings and offered limited access to a small group of AI safety researchers, but has not announced public weights or an API.
Does the announcement prove Dessa can clone a person’s voice?
No. The source report says Block has not clarified whether Dessa can imitate a specific person. Its demonstrations do not establish that capability.
Why are fraud concerns part of the discussion?
More natural-sounding speech could make it harder to recognize an impersonation call. The source report cites FBI data for 2025, including losses from distress scams involving cloned voices; those figures are not linked to Dessa.
How can people check a suspicious call?
Do not rely on a familiar voice alone. Hang up and call back using a number already saved or independently verified; families may also agree on a private code phrase for urgent requests.
Source: ThorstenMeyerAI.com
