Block’s Dessa Laughs, Hesitates, And Interrupts. Why Voice Isn’t Proof

🔍 Read the full analysis: Block’s Dessa Laughs, Hesitates, And Interrupts. Why Voice Isn’t Proof on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get privacy and security gear delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Block has published demonstrations of Dessa, an unreleased speech-to-speech research model that the company says can listen and speak at the same time, respond to emotion, and interrupt naturally. The recordings show laughter, pauses and conversational back-and-forth, but no outside researcher or reporter has tested the model directly, and Block has not published a paper, benchmarks or public access. The distinction matters as regulators and consumers confront the risks of convincing voice impersonation.

Block announced Dessa on October 23 through a project page. The company says the model was trained from scratch on recorded human conversation, without text or pretrained models. It also says Dessa runs faster than real time on a MacBook Pro M4 Max or better. Block has not provided a latency figure, model size or benchmark results to substantiate those performance descriptions.

The company’s selected recordings include Dessa laughing mid-sentence, helping a speaker recall a movie title and responding to an anniversary surprise. Block says the clips are live conversations that were not steered. Co-founder Jack Dorsey described talking with the model as feeling like talking with a person, citing its pauses, emotion and cadence. Dessa lead Ragavan Thurairatnam called the work “a GPT3 moment for speech.” These are company statements about a system that remains unavailable for general testing.

Coverage of the recordings has been more qualified. The Verge said some exchanges were “a little awkward and uncanny,” while finding Dessa more natural-sounding than another model featured in a comparison. TechCrunch reported that it was unclear whether emotional expression was guided by tags. Block has opened a waitlist and says a small group of AI safety researchers will receive early access; it has announced neither released model weights nor an API.

When a Familiar Voice Is Not Proof

Dessa’s demos do not establish that it can impersonate a particular person or be used in a fraud. They do illustrate why natural pauses, laughter and interruptions are weak identity checks: listeners may treat those cues as evidence that a caller is familiar, even though generated speech can reproduce conversational patterns. The risk is not unique to Block’s model, and the demonstrations alone do not show how often such deception succeeds.

The source material cites FBI figures from the 2025 Internet Crime Report, released in April 2026: 22,364 complaints mentioned AI, with $893 million in reported losses. The report also recorded more than $5 million in losses to distress scams involving cloned voices of loved ones. Those figures describe reported complaints and losses, not proof that Dessa or any particular model caused them. They place the discussion in an existing fraud environment rather than a hypothetical one.

For people and organizations, the practical implication is to avoid relying on voice alone when authorizing payments or responding to urgent requests. Callbacks to a previously known number and agreed family verification steps provide checks outside the incoming call. Humanlike voice quality can make a conversation more engaging, but it cannot establish who is speaking.

What Block Has and Has Not Released

Block, the company behind Square and Cash App, says Dessa comes from a small research lab formed by former members of the Toronto deep-learning startup Dessa, which Square acquired in 2020, according to CoinDesk. Block frames the project as an effort to build speech systems that retain emotional signals that conventional speech-to-text pipelines may discard.

As of October 26, 2026, public evidence consists of Block’s project description and selected audio recordings, along with journalists’ assessments of those clips. There is no public technical paper, model card, license or benchmark suite in the supplied reporting. Block says the model chooses its own voice and personality, but that does not answer whether it can reproduce a specific person’s voice.

The policy context is also moving. The source article says the European Union AI Act’s Article 50 provisions apply from August 2, 2026, including requirements relevant to disclosure of certain AI-generated or manipulated content. The exact obligations depend on the use and circumstances; the source does not establish how any specific Dessa deployment would be treated.

“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 Are Still Missing

Key technical and safety questions remain unanswered. Block has not released model weights or an API, and no outside party has reported running Dessa directly. There is no published information in the supplied material about its model size, measured latency, language coverage, training data documentation, watermarking or consent checks. Block has also not said whether the system can imitate an identifiable person.

It is not clear how much of the emotional behavior comes from the underlying model, how the demonstrations were selected, or whether prompts or other controls shaped the results. The recordings show what Block chose to present; they do not establish reliability across users, languages, noisy calls or adversarial situations. Nor do they establish that the system is available to scammers or that it has been used in fraud.

Researchers Await Early Access

Block says a small group of AI safety researchers will receive early access, but it has not announced a public release date. Independent testing and technical documentation would help clarify how the model performs beyond the published demonstrations, what safeguards it includes and whether its claims about local operation and speed hold under measured conditions.

Until those details emerge, people should treat an incoming voice as unverified when a caller asks for money, credentials or urgent action. Verify through a separate, trusted channel rather than relying on how familiar the voice sounds. For organizations, payment approvals can require callbacks to established numbers and written checks; families can agree on a code word for emergencies. Whether Block will publish further research or safety details 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 once and respond to conversational cues such as emotion and interruptions. It is not publicly available.

Has anyone independently tested the model?

Not according to the supplied reporting as of October 26, 2026. Coverage has assessed Block’s recordings, but there is no published paper, benchmark or outside report from running the model.

Does Dessa show that it can clone a person’s voice?

No. The demonstrations do not establish that Dessa can imitate a specific person. Block has not publicly clarified that capability, and it has not disclosed details about consent checks or watermarking.

How can callers reduce the risk of voice scams?

Do not treat a familiar-sounding voice as proof of identity. Hang up and call a number already known to be legitimate; families can also agree on a private code word for urgent requests.

Source: ThorstenMeyerAI.com