Why Block’s Dessa’s Human-Like Voice Leaves The Bigger Question Open

🔍 Read the full analysis: Why Block’s Dessa’s Human-Like Voice Leaves The Bigger Question Open on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Block announced Dessa, a research model that generates spoken conversation and, according to the company, listens while it talks. The October 23 demonstrations show laughter, pauses and interruptions, but the model has not been released for public or independent testing, so its performance, limitations and protections against voice impersonation remain unknown.

Block, the company behind Square and Cash App, says Dessa was trained from scratch on recorded human conversation alone, without text or pretrained models. The company describes it as a speech-to-speech system that can process a speaker’s words and vocal cues while responding, including through interruptions and short acknowledgments. Block says it runs faster than real time on a MacBook Pro M4 Max or better, but has not published a latency figure.

The company’s project page includes recordings of Dessa laughing mid-sentence, helping a speaker recall a film title and responding to an anniversary surprise. Block says each clip is a live conversation and that nobody steered the model. These recordings are company-selected demonstrations, not an independent evaluation. Block has not published a technical paper, model card, benchmark results, license or release date, and it has not announced public access or an API.

Outside coverage has offered a more qualified assessment. The Verge described some exchanges as “a little awkward and uncanny,” while saying Dessa came across as more natural than another model in a comparison. TechCrunch reported that it was unclear whether emotional expression was being guided by tags. Block’s research lead, Ragavan Thurairatnam, called the work “a GPT3 moment for speech” and said it could run on consumer GPUs; those are his assessment and the company’s claims, not independently verified results.

Natural Speech Raises Fraud Risks

Dessa’s appeal and its risk stem from the same feature: speech that may sound less scripted. A system that handles pauses, emotional cues and interruptions could make voice interfaces more comfortable to use. If similar systems can convincingly reproduce conversational behavior, however, it may become harder for a caller to judge whether a voice belongs to a relative, colleague or customer-service agent.

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. It also says victims lost more than $5 million to distress scams involving cloned voices of loved ones. Those figures describe reported complaints and losses across AI-related crimes; they do not establish that Dessa was used in any scam or that its demos make fraud more likely.

The practical implication is that a familiar voice alone may be a weak way to verify a request for money or sensitive information. Businesses and families can rely on previously established callback numbers and other verification steps rather than trusting caller ID or a convincing voice. The source article also recommends family code words and treating pressure to act urgently as a warning sign.

A Research Demo, Not a Public Product

Block published the Dessa project page on October 23, 2026. The team is described as a small research lab at Block, founded by former members of the Toronto deep-learning startup Dessa, which Square acquired in 2020. Block says the model was trained using one NVIDIA GB200 NVL72 and learned from far fewer words than large language models typically require. The company has not provided enough technical detail for outside observers to verify those comparisons.

Access remains limited: according to the source material, there is a waitlist and early access for a small group of AI safety researchers, but no public release or release date. The company says Dessa chooses its own voice and personality. That statement does not answer whether the system can imitate a particular person, how it handles consent or whether generated audio is marked so listeners can identify it.

Block argues that emotion-aware speech could help voice systems respond more appropriately when people are distressed. Critics cited in the source material question whether human-like voices make it easier to lower a listener’s guard. Both concerns are about the possible use of natural-sounding speech; neither, by itself, establishes how Dessa behaves outside the published recordings.

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

— Block co-founder Jack Dorsey

Capabilities and Safeguards Remain Untested

As of October 26, 2026, no independent researcher or reporter has been able to run the model, according to the source material. Without public access, published benchmarks or a technical paper, it is not possible to establish how consistently Dessa handles interruptions, different speakers, noisy conditions or conversations beyond the selected examples. Its language coverage and performance in languages other than those heard in the demos are also unclear.

Block has not disclosed whether the model can imitate a specific person, what consent checks apply, or whether generated speech is watermarked or otherwise identifiable. The material does not establish whether the model will be released beyond the limited researcher access, when that might happen, or what safeguards would accompany any wider access. No evidence provided links Dessa itself to fraud or shows that its recordings were made without consent.

Access and Technical Details to Watch

The next concrete developments would be a technical paper or model card, a public evaluation, and details about access and safety controls. Those could help researchers assess the company’s claims about training, speed and conversational ability. Block has not announced a timetable for any of those steps, so the timing of outside testing remains unknown.

Until independent testing is possible, the demos should be treated as illustrations of what Block says it has built, not proof of broad reliability or a settled risk profile. Organizations handling payments or sensitive requests can review caller-verification procedures now; families can agree on a callback routine or code word without assuming that any particular voice model has been used against them.

Key Questions

What is Block’s Dessa?

Dessa is a research speech-to-speech model that Block says was trained from scratch on recorded human conversation. The company says it can listen and speak at the same time.

Can the public test the model?

Not broadly, based on information available as of October 26, 2026. The source material describes a waitlist and limited access for some AI safety researchers, with no public release date or API announced.

Has Dessa been independently verified?

No independent testing is reported. Block’s published recordings are company-selected demos, and the company has not released a technical paper, model card or benchmark results.

Does the announcement show that Dessa is being used for scams?

No. The source material provides no evidence that Dessa has been used in fraud. It discusses broader concerns about voice cloning and cites FBI figures on reported AI-related crime.

What should people do about suspicious voice requests?

For requests involving money or sensitive information, verify through a number already known to you or another established channel. Families may also agree on a private code word and pause when a caller demands urgent action.

Source: ThorstenMeyerAI.com