🔍 Read the full analysis: Block’s Dessa And The Limits Of Judging AI By Its Voice 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 speak at once while conveying emotion. The public evidence consists of company-selected demos: there is no published paper, benchmark, model card or public release. Its humanlike voice also sharpens concerns about voice-based fraud, though the source material does not show that Dessa has been used in scams.
Block announced Dessa, a speech-to-speech research model, on October 23, presenting recordings in which it laughs, responds to emotion and speaks over conversational pauses. The demonstrations have drawn attention to the model’s natural-sounding voice, but no outside researcher has tested it, and Block has not published a technical paper, benchmarks or a public release date.
Block says Dessa was trained from scratch on recorded human conversation, without text or pretrained models. The company describes it as full-duplex: it listens while speaking and can interrupt or provide conversational acknowledgments. Its project page says it can handle features such as sarcasm, laughter and changes in emotional tone. These are Block’s descriptions, not independently measured results.
The company also says inference runs faster than real time on a MacBook Pro M4 Max or better, and that training used one NVIDIA GB200 NVL72. Block has not provided a latency figure, disclosed the model’s size or released benchmark results. The recordings include laughter during a sentence, help recalling a film title and a response to a surprise. Block says the clips are live conversations that were not steered; outsiders have not been able to reproduce or independently inspect them.
Access remains limited to a waitlist and early access for a small group of AI safety researchers, according to the source report. Block has not announced public model weights or an API. The Verge described some exchanges as occasionally awkward and uncanny, while finding Dessa more natural-sounding than another model in a comparison. TechCrunch reported uncertainty over whether emotional effects might be guided by tags.
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 Raises Fraud Risks
Dessa matters beyond the quality of a demonstration because humanlike voice can make impersonation harder to recognize. That does not establish that Dessa has been used for fraud, or that it can imitate a particular person. But the source report cites FBI figures showing 22,364 complaints mentioning AI in 2025, with $893 million in reported losses. The FBI also counted more than $5 million in losses from distress scams that use cloned voices to mimic loved ones, according to the report.
Those figures describe complaints and reported losses across AI-related fraud; they do not measure Dessa’s impact. The broader concern is that conversational cues—pauses, interruptions and emotional responses—can make a call sound more credible. If voice systems become more natural, a familiar-sounding voice alone is a weaker basis for trust. For businesses and families, independent verification matters more than how convincing a caller sounds.
The source material recommends using a known number to call back before acting on requests involving money, and setting a family code word for emergencies. For organizations, it points to callbacks for payment changes and written consent for synthetic voices. These are risk-reduction measures, not proof that a particular model can defeat existing safeguards.
Products Worth Considering
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
As an affiliate, we earn on qualifying purchases.
What Block Has Actually Released
Dessa comes from a small research lab at Block, the company behind Square and Cash App. The source report says the lab was founded by former members of Toronto-based deep-learning startup Dessa, which Square acquired in 2020. Block co-founder Jack Dorsey described talking with the model as feeling like talking with a person, citing its emotion, nuance, pauses and cadence. Lab lead Ragavan Thurairatnam called it “a GPT3 moment for speech” and said it could run on consumer GPUs.
Those comments are company and team characterizations, not the outcome of a published evaluation. Block says its approach learns from conversation rather than text and presents that as a different path for speech AI. The public material does not yet establish how well the model performs across users, settings or languages, or how it compares on controlled tests with other systems.
The response to the demos reflects the two sides of realistic speech. Some viewers praised how human it sounded; filmmaker Justine Bateman questioned the purpose of making AI voices sound human, while Block has argued that emotional awareness could help voice AI respond appropriately to people. The same quality can support more natural interaction and make synthetic speech more persuasive. The demonstrations establish what Block chose to show; they do not settle how the model behaves generally.
Key Tests Have Not Been Published
As of October 26, there is no public paper, model card, benchmark suite, license or confirmed release date in the source material. No outsider has run the model. That leaves basic questions open: how reliably it handles interruptions, whether its emotional responses are steered by tags, how it performs across languages, and what latency or hardware requirements apply beyond Block’s description.
Block has not publicly detailed watermarking, consent checks or safeguards against requests to imitate a specific person. The company says Dessa selects its own voice and personality, but that does not answer how the system would behave under other conditions. The available recordings cannot establish general reliability, safety, or whether the model can be used to impersonate someone.
The source report also cites the EU AI Act’s Article 50 as applying from August 2, 2026, in relation to disclosure of AI-generated content. The exact obligations for a given product or use are not detailed in the material, so readers should not treat the date alone as a complete account of compliance requirements.
Independent Evaluation Is the Next Test
The next meaningful developments would be publication of a technical paper or model card, access for independent evaluators, and clearer information about benchmarks, languages, safeguards and release plans. Block has said a small group of AI safety researchers has early access, but the source material gives no timetable for public testing.
Until those details emerge, the demos should be treated as examples selected by the company, not proof of broad performance. Organizations handling payment requests can verify changes through established contact channels, while families can agree on a code word and call back using a number they already trust. Those steps address voice impersonation generally; no evidence provided links Dessa itself to a fraud incident.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block announced on October 23, 2026. The company says it can listen and speak at the same time and was trained on recorded human conversation.
Has anyone outside Block tested the model?
Not according to the source material as of October 26. Public reporting has reviewed Block’s recordings, but no outside researcher has run the model, and no benchmark results or technical paper have been published.
Can Dessa clone a specific person’s voice?
The available material does not establish that. Block says Dessa selects its own voice and personality, but it has not publicly detailed whether the model can imitate a specific person or what safeguards apply.
Does the announcement show that Dessa has been used in scams?
No. The source material discusses AI-related fraud risks and FBI complaint figures, but does not report that Dessa has been used in a scam. The fraud figures cover broader AI-related complaints and losses.
When will the public be able to use Dessa?
Block has not announced a public release date, model weights or an API. Access is described as limited to a waitlist and a small group of AI safety researchers.
Source: ThorstenMeyerAI.com
