Block’s Dessa Interrupts Like A Person, But Does That Prove Anything?

🔍 Read the full analysis: Block’s Dessa Interrupts Like A Person, But Does That Prove Anything? on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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

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

As an affiliate, we earn on qualifying purchases.Hide

Block announced Dessa, a speech-to-speech research model, on October 23, showing recordings in which it laughs, pauses and interrupts in conversation. The company says the model was trained from scratch on recorded human conversation and can listen while speaking, but the demonstrations have not been independently verified: there is no public model, technical paper or benchmark, leaving its real-world performance unknown.

Block describes Dessa as a speech-only system, trained without text or pretrained models. It says the model can recognize and produce cues such as laughter, sarcasm and pauses, and can backchannel or cut in while a person is speaking. The company also says inference runs faster than real time on a MacBook Pro M4 Max or better, but has not published a latency measurement.

The recordings on Block’s project page include a conversation about a forgotten film title, laughter during a sentence and a response to an anniversary surprise. Block says each clip is a live recording and that nobody steered the model. Those are company-selected examples, not results from an outside evaluation. The Verge described some exchanges as awkward or uncanny, while finding Dessa more natural-sounding than another model featured in a comparison.

Block has not released model weights or an API. As of October 26, access is through a waitlist, with early access for a small group of AI safety researchers, according to the source material. The company has not announced a public release date, and key technical details, including model size and benchmark results, remain undisclosed.

Realistic Voices Complicate Call Checks

Dessa’s demos matter less as proof of a new capability than as a reminder that natural-sounding speech can make voice-based trust harder. A system that responds fluidly, handles interruptions and uses emotional cues could make an automated caller seem more like a person. That does not establish that Dessa can impersonate a specific individual or is being used for fraud; neither point has been demonstrated publicly.

Voice cloning is already part of reported fraud. The FBI’s 2025 Internet Crime Report, published in April 2026, counted 22,364 complaints mentioning AI, with $893 million in reported losses. The report also recorded more than $5 million in losses from distress scams using cloned voices to mimic loved ones. These figures cover a range of AI-related complaints and do not show that Dessa caused or contributed to those losses.

For businesses and families, the practical implication is that a familiar-sounding voice should not stand alone as proof of identity. For sensitive requests, people can hang up and call a known number or use a prearranged family verification phrase. Businesses can require a second channel for payment changes and urgent transfers. These precautions address voice fraud generally; the source material does not establish a specific threat from Dessa.

A Block Research Demo, Not a Release

Block, the company behind Square and Cash App, posted the Dessa project page on October 23. Co-founder Jack Dorsey said, “Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.” That is Dorsey’s assessment of the experience, not an independent finding about the model’s abilities.

Block says Dessa was trained using one NVIDIA GB200 NVL72 and learned from far fewer words than large language models typically require. It has not provided enough technical information for outside researchers to assess those claims or compare them with other systems. The team is described as a small research lab at Block, founded by former members of the Toronto AI startup Dessa, which Square acquired in 2020.

Coverage has also highlighted unanswered questions. TechCrunch said it was unclear whether emotional expression might be steered by tags, while The Decoder cautioned that sounding emotional does not establish emotion or intelligence. Those observations do not prove the model is scripted or lacks conversational ability; they underscore that the available clips cannot settle those questions.

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

— Jack Dorsey, Block co-founder

Independent Testing Has Not Begun

It remains unclear how Dessa performs across ordinary conversations, noisy calls, different speakers or languages. Block has not published a technical paper, model card, benchmark results or license, and outside researchers have not publicly tested the model. There is also no disclosed list of supported languages or public release date.

The available evidence does not establish whether the system can imitate a particular person, what protections against impersonation may be included, or whether its emotional responses are generated without steering. Block says Dessa selects its own voice and personality, but that description does not answer questions about voice consent, safeguards or the limits of the system. The published clips cannot show how reliably it handles cases beyond the examples chosen for the page.

Watch for Access and Technical Evidence

The next meaningful developments would be access for independent researchers, published technical documentation and evaluations using a broader range of conversations. Those could help establish whether the model’s turn-taking and expressive speech are consistent, what hardware and latency it requires, and how its performance compares with other systems. No publication or release date has been announced.

Until then, readers should treat Block’s descriptions as company claims and its clips as demonstrations, not proof of general performance. If Block releases the model or new safety information, questions about consent, imitation, disclosure and misuse will be central to judging how it can be used responsibly.

Key Questions

What is Block’s Dessa?

Dessa is a speech-to-speech research model that Block says can listen and talk at the same time. The company says it was trained from scratch on recorded human conversation without text or pretrained models.

Has an independent group verified the demonstrations?

Not publicly, based on information available as of October 26, 2026. The recordings are selected by Block, and no outside researcher has publicly reported testing the model.

Can Dessa clone a specific person’s voice?

The available information does not establish that. Block says Dessa chooses its own voice and personality, but has not publicly detailed voice-imitation capabilities or related safeguards.

Is Dessa publicly available?

No public weights or API have been announced. Block’s project page offers a waitlist and, according to the source material, early access to a small group of AI safety researchers; a public release date is unknown.

What can people do about voice-call scams?

Do not rely on a familiar voice alone for identity checks. Hang up and call a number you already know, use a family verification phrase, and confirm urgent payment requests through a separate trusted channel.

Source: ThorstenMeyerAI.com