🔍 Read the full analysis: AI Voice Has Changed: Why Sounding Human No Longer Settles Anything on ThorstenMeyerAI.com
Get hardware and tech essentials 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 on October 23, describing it as a speech-to-speech research model trained on recorded human conversation. Its demos show expressive, interruptible conversation, but the model has not been independently tested or publicly released, leaving its capabilities and safeguards unclear.
Block announced Dessa, a speech-to-speech research model that the company says can listen and speak at the same time, on October 23. Its company-produced recordings feature laughter, pauses and interruptions that sound conversational, but no independent tester has run the model and it is not publicly available, leaving its real-world performance and fraud risks unverified.
Block, the company behind Square and Cash App, says its small research team trained Dessa from scratch on human conversation alone, without text or pretrained models. The company says the system recognizes and produces emotion, including sarcasm, laughter and changes in vocal tone. It also says Dessa can backchannel and interrupt rather than waiting for a speaker to finish.
Block describes the model as running faster than real time on a MacBook Pro M4 Max or better, but has not published a latency figure. The company says training used one NVIDIA GB200 NVL72 and far fewer words than large language models typically require. Those details remain company claims: there is no public technical paper, model card, benchmark, downloadable model or API.
The recordings on Block’s project page include a conversation about forgetting a film title and a response to an anniversary surprise. Block says the clips are live recordings without steering. The Verge reported that some exchanges felt awkward or uncanny, while TechCrunch noted that it was unclear whether emotion was guided by tags. Block has offered a waitlist and early access to a small group of AI safety researchers, not general access.
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
Why Human-Sounding Calls Matter
Dessa matters beyond its own release because natural conversational cues can make it harder for listeners to distinguish a machine from a person. That has legitimate uses: Block says emotional awareness could help a voice assistant respond appropriately when someone is distressed. But the same cues—pauses, laughter and quick replies—could make an impersonation more persuasive. Human-sounding speech is not evidence of a human caller, and a convincing voice does not establish identity.
Voice cloning is already part of reported fraud. The FBI’s 2025 Internet Crime Report, published in April 2026, recorded 22,364 complaints mentioning AI and $893 million in reported losses associated with those complaints. It also counted more than $5 million in losses from distress scams using cloned voices of loved ones. These figures describe complaints and reported losses, not a measure of Dessa’s impact; there is no evidence in the supplied reporting that the model has been used in a scam.
For organizations, the practical concern is whether staff treat a familiar voice as authorization for a payment or account change. For families, an urgent call that appears to come from a relative can prompt action before anyone checks independently. Verification needs to happen through a separate, trusted channel, not by asking the caller to sound more convincing.
Products Worth Considering
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Jackery Explorer 2000 v2 Portable Power Station with Two 200W Solar Panels
As an affiliate, we earn on qualifying purchases.
From Voice Cloning to Conversation
Earlier voice-cloning scams could rely on short recordings and a listener’s familiarity with a person’s voice. A conversational model could add behaviors that make an exchange feel less like a prerecorded message: responding to interruptions, reacting to emotion and filling pauses. That is the capability Block is demonstrating, though the demonstrations do not establish how reliably it works outside the selected clips.
Block co-founder Jack Dorsey described talking with Dessa as feeling like talking with a person, citing emotion, nuance, pauses and natural cadence. Project lead Ragavan Thurairatnam called the work “a GPT3 moment for speech,” according to the source material. Those are assessments by people involved with the project, not independent test results. Reporters who heard the recordings offered a more mixed picture, and the model remains inaccessible to outside evaluators.
The distinction matters: the current news is a research announcement, not a verified public product launch. Block has not disclosed model size, benchmark results, a release date or a complete list of supported languages. The company says Dessa chooses its own voice and personality, but the available information does not establish whether it can imitate a specific person or what protections might prevent that.
“a little awkward and uncanny”
— The Verge, in its October 23 report
What Independent Tests Must Establish
The central limitation is that outsiders have not tested Dessa directly. Without a paper, model card, benchmarks or public access, it is not possible to verify Block’s claims about training, speed, conversational ability or the amount of data used. Selected demonstrations can show what happened in those recordings, but not how the model performs across ordinary calls, difficult conversations or adversarial attempts.
Safeguards are also not clear. The source material reports no public statement from Block about watermarking, consent checks or measures against imitating a specific person. The supported languages, release timing and access conditions beyond the limited researcher group remain undisclosed. It is also unknown whether emotional expression is generated directly by the model or shaped through controls such as tags.
There is no reported evidence that Dessa has caused fraud or that it is available to scammers. The broader fraud figures describe reported incidents involving AI, not cases attributed to this model. Its potential misuse is a risk to examine, not a confirmed outcome.
Release Details and Safety Checks
The next useful evidence would be a technical paper or model card, independent evaluations and clear information about access, supported languages and safeguards. If Block expands access beyond its small group of safety researchers, outside testing could clarify whether the system performs as its demos suggest and how it responds to impersonation attempts. No public release date has been announced.
Meanwhile, people and organizations do not need to wait for Dessa to change basic verification practices. A caller’s voice should not be the sole basis for sending money, changing account details or sharing sensitive information. Call back using a number already on file, and use a family code word as one additional check. These steps address voice-impersonation risk generally; they are not evidence of a specific threat from Block’s model.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says can listen and talk at the same time. The company says it was trained from scratch on recorded human conversation.
Can the public use or independently test it?
Not based on the information available as of October 26, 2026. Block has announced a waitlist and early access for a small group of AI safety researchers, but no public weights or API.
Has Dessa been used in a voice scam?
No such use is reported in the source material. The FBI figures cited concern AI-related complaints and voice-cloning distress scams generally; they do not attribute those cases to Dessa.
How should people verify an urgent call asking for money?
Do not rely on voice recognition alone. Hang up and call the person back using a number you already know, and consider agreeing on a family code word. Treat pressure to act immediately as a reason to verify independently.
Source: ThorstenMeyerAI.com
