This Voice Isn't Human: Block's Dessa and the Family Safe Word

Dessa is a new voice AI from Block, Jack Dorsey's company, announced on 23 October 2026. It listens while it talks, laughs mid-sentence, says "yeah" while you're still speaking and knows when to cut in. Block says it learned all of that from recordings of people talking, with no text at all.

Block also says it runs on a laptop, a MacBook Pro M4 Max or better. You can't use it yet, though. It's a research preview, with no public download, API or price.

What you can do today is listen to Block's 21 sample conversations and test your own ear. And agree a family safe word, because AI voices on the phone are getting harder to pick out.

How do you keep your family safe from AI voice scams?

When you can't tell an AI voice from a person, three habits keep you and your family safe.

1. Agree a family safe word, in person rather than by text, and keep it off social media. Anyone who calls in a panic asking for money has to say it.
2. If a call asks for money, codes or passwords, hang up and call back on a number you already have for that person or company.
3. Treat urgency as the red flag. Your bank and your family can wait five minutes while you check.

What is Block's Dessa?

Dessa is a speech-to-speech model: audio in, audio out. Block's AI research lab trained it from scratch on human conversation alone, with "no text, no pretrained models". As far as Block knows, it's the first AI you can have a realistic live conversation with that was trained only on human speech.

Block says most voice AI turns your voice into text, writes a reply, then reads it out. Its point is that text throws away how you said something: the sigh, the sarcasm, the voice cracking. Dessa hears and speaks those directly. Block compares it to how babies learn to talk, by listening, long before they see a written word.

It's full duplex, so it listens and talks at the same time. Block says it replies faster than real time, so there's no awkward wait.

The lab behind it is small. It was founded by former members of Dessa, the Toronto deep learning startup, it's led by Ragavan Thurairatnam, and it reports to Jack Dorsey, who co-founded Twitter and Block. Block says it trained Dessa on a single NVIDIA GB200 NVL72, one rack of GPUs.

Does Dessa really sound like a person?

That's Block's claim. Jack Dorsey called Dessa "the most authentic full-duplex speech model ever built" and said talking with it "feels like talking with a person". TechCrunch's headline called it eerily human. Block hasn't published a listening test, though, so nobody has measured how often people can tell.

So test your own ear. Block says its clips are live recordings of a person talking to Dessa, with nobody steering it and subtitles added afterwards. The exceptions are two conversations with GPT-Live-1, OpenAI's voice model, where both voices are AI.

One caveat from TechCrunch: it's not clear whether anything, like the voice tags other models use, steers the emotion you hear.

Can you try Dessa yet?

No. Block calls Dessa a research preview. There's no public download, API or price, and on 26 October the Download button on its page said "coming soon".

Block is giving a small group of AI safety researchers early access first, to find the risks it hasn't anticipated. If that's your field, there's an application on Block's page.

For everyone else, the 21 conversations are on Block's page, and you can download any of them.

If you're weighing up an AI voice for your own business line, see how an AI receptionist works, and read what the EU AI Act says about telling callers they're talking to AI.