Jack Dorsey’s financial services firm Block has introduced a voice AI system that displays strikingly human-like qualities during demonstrations, triggering reactions that oscillate between deep fascination over its communication capabilities and serious warnings about how such innovation could facilitate harder-to-spot phone frauds.
Known as Dessa, this model was constructed entirely from human speech data instead of textual input and possesses the ability to listen simultaneously while talking, permitting it to manage interruptions, chuckles, and shifts in vocal inflection mid-conversation, per the official announcement from Block.
Conversing with it feels identical to talking with an actual person, brimming with affect, subtlety, pauses, and organic rhythm, Dorsey noted on X.
During one particular showcase, Dessa giggles right in the middle of an utterance; in another, it responds to an unexpected moment with what appears to be authentic feeling. Block emphasized that these dialogues were captured live without any external scripting dictating the software’s replies.
Unlike conventional architectures that first transform spoken words into text prior to formulating a verbal reply, Dessa interprets sound waves directly. Block stated that it trained Dessa completely from the ground up utilizing a single Nvidia GB200 NVL72 computing cluster. Furthermore, the system is lightweight enough to execute locally on a MacBook Pro powered by an M4 Max processor or higher, according to the corporate release.
At present, Dessa exists strictly as a research preview rather than a commercially available tool.
Shock and awe
Dessa’s showcase swiftly captured widespread attention across social networks, with numerous commentators applauding how realistic the exchanges sounded.
Platform user Alexander Hoff remarked that he had never encountered anything synthetic that felt so genuinely human, whereas user Hesamation characterized the breakthrough as pure science fiction.
Binance founder Changpeng CZ Zhao also shared his perspective, highlighting potential applications within the educational sector, where organic verbal engagement could assist youngsters in learning through dialogue.
This might prove highly beneficial for GiggleAcademy, as educating youth using existing artificial voices presented difficulties, Zhao posted on social media, alluding to his schooling initiative.
Yet, those exact lifelike attributes that captivated specific observers also provoked worries among others regarding the system’s capacity for misuse.
Movie director Justine Bateman questioned the necessity for artificial voices to mimic humans at all, contending that such capabilities might be exploited to deceive individuals.
So you can trick people into lowering their guard, she posted.
Reporter Shiv Aroor voiced identical anxieties concerning criminal deception, observing that fraudulent telephone calls are poised to reach pandemic proportions.
Dessa originates from Block, the parent organization behind Square, Cash App, the Bitkey cryptocurrency wallet, and the Proto bitcoin mining hardware division. Its artificial intelligence research division reports directly to Dorsey and incorporates former employees originating from the Toronto-based AI enterprise Dessa, which Square purchased back in 2020.
Block has refrained from publishing the architecture size, performance metrics, or any timeline for a general rollout. Its audio demonstrations remain internal previews rather than independently verified evaluations.
The corporation is currently providing initial access exclusively to a select circle of artificial intelligence safety specialists. Block noted that speech-driven AI might eventually excel at detecting emotional indicators—such as suffering and anguish—that written transcripts frequently overlook, while investigators evaluate potential hazards that the enterprise might have failed to spot.
Originally published at https://www.coindesk.com/tech/2026/10/24/block-s-new-voice-ai-sounds-eerily-human-sparking-awe-and-scam-fears.
