🔍 Read the full analysis: Block’s Dessa Mimics Human Conversation. What Does That Actually Prove? on ThorstenMeyerAI.com
Get smart everyday buys 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 natural-sounding exchanges, but outsiders have not tested the model and Block has not published a paper, benchmarks or public access details. The demonstrations raise questions about both voice interfaces and the risks of convincing voice impersonation.
Block announced Dessa, a speech-to-speech research model that the company says can handle interruptions, laughter and other conversational cues, on October 23. The demonstrations offer a glimpse of more natural voice AI, but the model has not been independently tested, and Block has not released a technical paper, benchmarks or public access to it.
Block says Dessa was trained from scratch on human conversation audio, without text or pretrained models. The company describes it as full-duplex: it listens while it speaks, and can cut in or respond with conversational signals such as backchannels. Block also says it processes speech faster than real time and runs locally on a MacBook Pro M4 Max or better. It has not provided a latency figure.
The examples on Block’s project page include a voice laughing mid-sentence, helping a speaker recall a film title and reacting to news of an anniversary surprise. Block says the recordings are live conversations and that nobody steered the model. These clips are company-selected demonstrations, not a reproducible evaluation or evidence of how the system performs across users and situations.
Block co-founder Jack Dorsey described the experience as conversation with a person, citing emotion, nuance, pauses and natural cadence. Dessa lead Ragavan Thurairatnam called the work “a GPT3 moment for speech” and said it could run on consumer GPUs. Those are assessments from the company’s leadership; independent access has not been reported.
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 and Fraud Risks
A more fluid voice interface could make AI assistants easier to talk with, especially when users interrupt, hesitate or express emotion. Block argues that current voice systems can miss emotional cues, such as a caller being in pain. Dessa’s demos suggest the company is exploring a system that responds to more than the literal words in a conversation, though its practical performance is not yet established.
The same conversational realism could complicate efforts to identify impersonation. Voice fraud already exists; a model does not need to match Dessa’s capabilities to be used in a scam. The FBI’s 2025 Internet Crime Report, published in April 2026, counted 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 cover AI-related complaints and reported losses; they do not show that Dessa was involved.
The risk is not proof that this particular research model can clone a specific person or is being used for fraud. Rather, natural timing and emotional responses could make a synthetic caller harder to judge by voice alone. Businesses and families should rely on verification beyond a familiar-sounding voice, particularly when a call creates urgency or requests money.
Products Worth Considering
HP Sprocket Panorama Portable Color Photo and Label Printer
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
As an affiliate, we earn on qualifying purchases.
What Block Has Released So Far
Block, the company behind Square and Cash App, presented Dessa through a project page on October 23. The work comes from a small research lab at Block, which the source material identifies as founded by former members of Toronto deep-learning startup Dessa, acquired by Square in 2020. The shared name refers to the research project and the earlier startup; the available material does not describe a separate consumer product launch.
Block says the model was trained using “only one NVIDIA GB200 NVL72” and learned from hundreds of times fewer words than large language models require. These are company claims, and the word-count comparison is not accompanied by a published methodology. Block has not disclosed the model’s size or benchmark results. Its project page says access is currently limited to a waitlist and a small group of AI safety researchers.
Outside coverage has noted limits in the demonstrations. The Verge described some exchanges as occasionally awkward and uncanny, while finding Dessa more natural-sounding than a comparison model in the clips it reviewed. TechCrunch reported that it was unclear whether some emotional expression was guided by tags. Such observations concern selected recordings, not independent testing of the model itself.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
— Block co-founder Jack Dorsey
What Independent Tests Could Show
No outside researcher or reporter has publicly run Dessa, according to the source material available as of October 26. There is no technical paper, model card, benchmark suite, published license or confirmed public release date. Without those materials, claims about its training, performance, compute requirements and general reliability cannot be checked independently.
Other open questions include which languages it supports, how it performs in noisy or difficult conversations, and whether emotional behavior is generated directly or guided by tags. Block says the model chooses its own voice and personality, but the available information does not establish whether it can imitate a specific person, what safeguards would prevent that, or whether outputs are watermarked. It is also unclear how the model handles consent for recorded voices or what safeguards would accompany wider access.
The demos cannot establish whether the system is emotionally aware, consistently conversational or safe for use in customer service and other high-stakes calls. Nor is there evidence in the supplied material that Dessa has been used in fraud. Those distinctions matter: a persuasive demonstration is evidence of what Block chose to show, not a full account of the model’s capabilities or risks.
Release Details and Safety Measures
The next meaningful step is for Block to provide independent researchers with access and publish technical documentation, evaluation results and release terms. Those disclosures could clarify the model’s performance, limits, supported languages, safeguards and whether its speech can be identified as synthetic. Block has not announced a public release date, so when broader testing will be possible remains unknown.
Until then, callers and organizations should not treat a familiar voice as proof of identity. A call requesting money can be ended and verified using a number already on file; families can agree on a code word for urgent situations. Companies can require callbacks or a second channel to approve payments. These are general precautions against voice impersonation, not evidence of a specific threat from Dessa.
Any deployment would also make disclosure and consent questions more immediate. The source material says the EU AI Act’s Article 50 applies from August 2, 2026, but does not provide a legal analysis of how the provision applies to a particular use of Dessa. Block has not detailed a policy for synthetic-voice disclosure or consent. The model’s next stage will depend on whether the company supplies more evidence and whether outside testing confirms the capabilities shown in its clips.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says was trained from scratch on human conversation audio. The company says it can listen while speaking and respond to conversational cues such as interruptions and laughter.
Has anyone independently tested Dessa?
Not publicly, based on the information available as of October 26, 2026. Reporters have reviewed Block’s demos, but no outside party has publicly run the model.
Can Dessa clone a particular person’s voice?
That is not established. Block says the model selects its own voice and personality, but has not publicly explained whether it can imitate a specific person or what safeguards would prevent impersonation.
Is there evidence that Dessa has been used in scams?
The supplied information does not show that Dessa has been used in fraud. The FBI’s reported AI-related complaint and loss figures describe broader activity and should not be attributed to this model.
When can the public use Dessa?
Block has not announced a public release date. The company says access is limited to a waitlist and a small group of AI safety researchers.
Source: ThorstenMeyerAI.com
