Why Block’s Dessa’s Conversational Voice Can’t Prove Who’s Speaking
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Why Block’s Dessa’s Conversational Voice Can’t Prove Who’s Speaking on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

Block announced Dessa, a speech-to-speech research model whose demonstrations feature laughter, interruptions and emotional-sounding responses. The recordings are company-selected, and outsiders have not tested the model; its release, safeguards and ability to imitate a specific person remain unclear. Its natural conversational style matters because voice cloning is already used in fraud, but sounding human does not prove who is speaking.

Block has announced Dessa, a speech-to-speech research model that can respond with laughter, pauses and interruptions in company demonstrations, but no independent testing has established how reliably it works or whether it can verify a speaker’s identity. The distinction matters as fraudsters already use cloned voices in impersonation schemes: conversational realism may make a synthetic voice harder to question, but it is not proof of who is on the line.

Block published a project page for Dessa on October 23. The company says it trained the model from scratch on recorded human conversation, without text or pretrained models. It describes the system as full-duplex, meaning it can listen while speaking and can interrupt or respond during another speaker’s turn. Block also says it can handle vocal cues such as sarcasm, laughter and sighs.

The demonstrations on Block’s page include a model laughing mid-sentence, helping a speaker recall a film title and responding to an anniversary surprise. Block says the clips are live recordings and were not steered. Those are company-provided examples, not an independent evaluation. The Verge described some exchanges as occasionally awkward and uncanny, while also finding Dessa more natural than the other model in one comparison. TechCrunch reported that it was unclear whether emotion in the responses was guided by tags.

Block says inference runs faster than real time on a MacBook Pro M4 Max or better, but has not supplied a latency figure. The company also says the model was trained using one NVIDIA GB200 NVL72. Block has not published model weights, an API, a technical paper, a model card, benchmark results or a public release date. Access is described as a waitlist and early access for a small group of AI safety researchers.

At a glance
reportWhen: Announced October 23, 2026; status as o…
The developmentBlock announced its Dessa conversational voice research model on October 23, with demonstrations that have raised questions about how natural-sounding AI could affect voice fraud.
Thorsten Meyer AI
Status · 26 Oct 2026

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.

1NVIDIA GB200 NVL72 rack used for training, per Block
Notext in the training data, per Block
0independent tests published so far
M4 MaxMacBook Pro class Block says runs it locally
TopicWhat we knowStatus
Training dataHuman conversation only: “no text, no pretrained models”Block’s claim
ConversationFull-duplex: listens while it talks, backchannels, cuts inBlock’s claim
EmotionHears and speaks laughter, sarcasm, sighs; unclear if steered by tags (TechCrunch)Block’s claim
Speed“Faster than real time”; no latency figureBlock’s claim
HardwareTrained on one NVIDIA GB200 NVL72; runs on a MacBook Pro M4 Max or betterBlock’s claim
DemosSome chats “a little awkward and uncanny”, more natural than GPT-Live-1 (The Verge)reporters heard it
Size, benchmarks, release dateNot disclosed (CoinDesk)not disclosed
Paper, model card, license, weightsNone found as of 26 Oct 2026not disclosed
LanguagesNot listed; Block calls learning new languages an open questionnot disclosed
Voice cloning safeguardsNo statement on watermarking, consent checks or imitating a specific personnot 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 Natural Voices Complicate Call Security

A voice that responds fluidly can make a call feel more credible, but natural speech is not identity verification. Dessa’s conversational features do not demonstrate that it can confirm who is speaking, nor do the public demonstrations establish whether it can reproduce a particular person’s voice. Those questions are separate from whether its responses sound convincing.

The risk is not hypothetical at the level of voice cloning generally. The source material 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 using cloned voices to imitate loved ones. These figures cover reported complaints and losses; they do not show that Dessa was used in any scam.

For organizations, the practical implication is to avoid treating a familiar voice as sufficient authorization for payments, account changes or sensitive disclosures. Independent callbacks to a known number and other verification steps reduce reliance on the voice alone. For families, agreeing on a private code word and calling back using a saved number can help check an urgent request. These precautions address voice impersonation broadly, not a demonstrated use of Dessa.

Ninja Air Fryer XL, 6.5 Qt, 6-in-1, MaxCrisp, Dark Grey (AF181)

Ninja Air Fryer XL, 6.5 Qt, 6-in-1, MaxCrisp, Dark Grey (AF181)

As an affiliate, we earn on qualifying purchases.

What Block Has Disclosed About Dessa

Block is the company behind Square and Cash App. Its small research group describes Dessa as a speech-focused model rather than a conventional text-based assistant with speech added on. The project name also relates to Dessa, a Toronto deep-learning startup acquired by Square in 2020, according to CoinDesk.

Block co-founder Jack Dorsey wrote that talking with the model feels like talking with a person and praised its emotion, nuance, pauses and 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 by people involved in the project, not outside findings about performance.

The company frames its work as a way for voice systems to respond to emotion that speech-to-text pipelines can lose. That potential use and the concern about impersonation arise from the same feature: a system that attends to conversational cues may feel more responsive. Block’s demonstrations establish what it chose to show, but do not settle how the model performs across speakers, situations or languages.

“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”

— Block co-founder Jack Dorsey

Independent Tests and Safeguards Are Missing

As of October 26, no outside researcher or reporter is reported to have run the model. There is no public paper, model card, benchmark, license or release date in the source material. Without access to the system and technical documentation, its accuracy, limitations, latency and performance outside the selected demonstrations cannot be independently assessed.

It is also unclear which languages Dessa supports, how it handles different accents and noisy calls, and whether its emotional responses rely on labels or other steering. Block says the model chooses its own voice and personality, but the available material does not answer whether it can imitate a particular person. Details about watermarking, consent checks and other safeguards have not been disclosed in the source material.

Nothing in the demonstrations establishes that Dessa has been used for fraud, that it can defeat existing verification systems, or that it can reliably identify the person speaking. The model’s conversational ability and its capacity to prove identity are different questions; the latter remains untested publicly.

Access and Documentation Will Matter

The next meaningful developments would be broader researcher access and technical documentation that allows independent evaluation. Block has described a waitlist and early access for a small group of AI safety researchers, but has not announced when a wider release, weights or API might become available. Any later claims about capability should be weighed against reproducible tests, not demonstrations alone.

Until more information is available, businesses can review call-based approval procedures and require a second, independently verified channel for high-impact requests. Families can establish a callback routine and a shared code word before an urgent call occurs. These steps do not depend on whether a particular model is involved; they account for the existing possibility of voice impersonation.

Further reporting should establish whether Block publishes safeguards or test results, how the model behaves across languages and real-world call conditions, and whether it can reproduce a specific person’s voice. For now, Dessa’s public evidence is limited to Block’s demonstrations, and no public release timeline has been given.

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 was trained on recorded human conversation and can listen while speaking.

Has anyone independently tested the model?

The source material reports that no outsider has run the model. Reporters have commented on Block’s recordings, but those comments are not independent tests of the system itself.

Does Dessa prove who is speaking?

No evidence provided so far shows that Dessa can verify a caller’s identity. A voice may sound natural or familiar without proving who produced it.

Can Dessa imitate a specific person?

That remains unclear. Block says Dessa chooses its own voice and personality, but the source material gives no test results or clear disclosure about whether it can imitate a particular person.

How can people check an urgent voice request?

Use a known callback number rather than the number on an unexpected call, and consider agreeing on a private family code word. Organizations should use separate verification steps for payments and sensitive account changes.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Thanksgiving Hosting Essentials: A Thanksgiving Guide

Plan your Thanksgiving menu, kitchen, budget, and guest space with practical food-safety tips and a calmer day-of schedule.

Dow

The Dow Jones Industrial Average dropped sharply today, reflecting increased market volatility amid economic concerns and corporate earnings reports.

The Second Trump Presidency Is A World-historic Failure

A Noahpinion analysis points to weak approval ratings and voter concerns as evidence that Trump’s second presidency has faltered so far.

Copper-hungry US Buyers Circle Ivanhoe Electric Mine

Ivanhoe Electric says it is negotiating copper sales for its $1.4 billion Santa Cruz project, with potential buyers and financing still unconfirmed.