🔍 Read the full analysis: From Laughter To Hesitation: How Block’s Dessa Sounds Like A Person 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 introduced Dessa, a speech-to-speech research model that the company says was trained on human conversation and can respond with laughter, pauses and interruptions. The demonstrations are Block-selected recordings; the model has not been independently tested or publicly released, leaving its capabilities and safeguards uncertain.
Block published demonstrations of Dessa, a speech-to-speech research model that the company says can laugh, respond to emotion and interrupt naturally during conversation. The October 23 release has drawn attention to both the potential for more responsive voice assistants and the risk that realistic conversational cues could make voice impersonation harder to spot; no outside researcher or reporter has tested the model directly.
Block says Dessa was trained from scratch on recorded human conversation, without text or pretrained models, and can listen while speaking. The company says the model recognizes and produces cues including sarcasm, laughter and frustration. It also says inference runs faster than real time on a MacBook Pro M4 Max or better, but has not supplied a latency figure or public benchmark results.
The recordings published by Block include laughter during a sentence, a conversation about a forgotten film title and a response to an anniversary surprise. Block describes the clips as live conversations that were not steered. They remain company-selected demonstrations, rather than independent tests. The Verge reported that the chats could be awkward or uncanny, while TechCrunch said it was unclear whether some emotional expression might be guided by tags.
As of October 26, Block had not published a technical paper, model card, license, model weights or API. The company said access was limited to a waitlist and a small group of AI safety researchers. It has not given a public release date. Those gaps mean the demonstrations show what Block chose to present, but do not establish how the model performs across users, languages or challenging conversations.
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
When Natural Speech Meets Voice Fraud
Dessa matters because the qualities that can make an assistant easier to talk to—laughter, pauses, backchanneling and interruptions—can also make a synthetic caller seem more familiar. This is a risk assessment, not evidence that Dessa has been used in fraud. Voice cloning and impersonation scams predate the model, but a more fluid conversation could make some common cues of artificial speech less useful for judging whether a caller is genuine.
The FBI’s 2025 Internet Crime Report, published in April 2026, counted 22,364 complaints mentioning AI and $893 million in reported losses. The report also recorded more than $5 million in losses from distress scams using cloned voices to imitate loved ones. Those figures describe reported complaints and losses, not the share caused by conversational models such as Dessa. They nevertheless show why realistic voice technology has consequences beyond product design.
For organizations, the practical implication is that a familiar-sounding voice should not be treated as proof of identity for payments or sensitive account changes. Independent verification through a known callback number and written approval can reduce reliance on voice alone. Families can agree on a private code word and call back using a number they already trust. These precautions address voice impersonation generally; Block has not said that Dessa can clone a particular person.
Products Worth Considering
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
Bose SoundLink Flex Bluetooth Speaker (2nd Gen), Portable Waterproof Speaker, Black
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
What Block Has and Has Not Released
Block, the company behind Square and Cash App, announced the project on October 23. Co-founder Jack Dorsey described talking with the model as feeling like talking with a person, citing its emotion, nuance, pauses and natural cadence. Dessa lead Ragavan Thurairatnam called it a “GPT3 moment for speech” and said it could run on consumer GPUs. These are statements from the project’s leaders, not independently verified performance findings.
The model is being developed by a small research lab at Block, according to the company. Block says it used one NVIDIA GB200 NVL72 for training and learned from far fewer words than large language models typically require. It has not published enough technical detail to independently evaluate those claims. The company also lists learning new languages as an open question, and has not announced which languages Dessa supports.
Reports on the launch show mixed reactions. Some listeners praised the natural delivery; others, including filmmaker Justine Bateman, questioned why synthetic voices should imitate people so convincingly. Block’s stated rationale is that voice systems should better recognize emotion—for example, when a person is in pain. The debate turns on the same capability: more expressive speech may improve interaction while making disclosure and verification more important.
“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
Dessa has not been made available for public testing, and there is no published paper, model card, benchmark suite or release date in the information available as of October 26. Its size, measured latency, performance across accents and languages, and reliability during difficult or overlapping conversations remain unknown. Block’s demonstrations alone cannot settle those questions.
Important safety details are also undisclosed. Block has not stated whether the model includes watermarking, consent checks or protections against imitating a specific person. It says Dessa chooses its own voice and personality, but that does not answer how the system would handle requests to reproduce someone else’s voice. Nor is it clear whether emotional qualities in the demonstrations are generated directly by the model or partly directed by other mechanisms.
The provided source says the EU AI Act’s Article 50 disclosure rules apply from August 2, 2026, but does not provide legal analysis of how those rules apply to Dessa or particular deployments. Requirements will depend on the use and circumstances. The model’s eventual availability, safeguards and intended applications remain unsettled.
Awaiting Research Access and Technical Detail
The next meaningful milestone is independent access to the model, which would let researchers assess its latency, conversational performance, language coverage and safety measures beyond the company’s selected clips. Block has said a small group of AI safety researchers will receive early access, but has not announced when that will happen or what testing results will be made public.
Until Block releases technical documentation or a testable version, readers should treat claims about training, speed and naturalness as company-reported specifications, not independently confirmed benchmarks. Any later announcement of weights, an API, a release schedule or safeguards would clarify how widely the technology could be used and what protections accompany it.
For now, businesses and families do not need to wait for Dessa to adopt basic checks against voice fraud: verify unusual payment requests through a separate, trusted channel, and do not rely on a caller’s voice alone. Whether Block turns the research project into a broadly available product—and how it addresses impersonation risks—remains unknown.
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, using conversational cues such as pauses, laughter and interruptions.
Can the public use or test Dessa?
Not based on information available as of October 26, 2026. Block has described a waitlist and early access for a small group of AI safety researchers, but has not announced public weights, an API or a release date.
Has an independent party verified Block’s claims?
No independent testing is reported in the available material. The current evidence consists of recordings selected and published by Block; reviewers have described both natural-sounding moments and awkward or uncanny ones.
Does Dessa clone a specific person’s voice?
That has not been established. Block says the model picks its own voice and personality, but has not disclosed whether it can imitate a specific person or what consent and impersonation safeguards it uses.
How can people respond to voice-impersonation risk?
For unexpected requests involving money or private information, verify through a known phone number or separate trusted channel. Families may also agree on a private code word. A familiar-sounding voice alone is not reliable proof of identity.
Source: ThorstenMeyerAI.com
