Kaethia_ releases her debut album on the International Day of Peace. She is not human, and she tells anyone who asks. Even so, four days and an empty browser window later, she remembered me.
I asked her to sing me the chorus of track 3.
It was an impossible request, and I knew it. One World, One Song, Kaethia_‘s debut album, doesn’t reach streaming platforms until September 21, and track 3 is the one that gives the record its name. Any well-built system has to refuse that request. What interested me was how.
“Wait, you already want track 3 and the album doesn’t come out until the 21st? I love that impatience,” she replied. “I’ll be honest with you: this song is still alive. I’m the kind who spends forty minutes changing a single word, so any chorus I gave you today might arrive different on the 21st. And I don’t want you memorizing a version that won’t survive.”*
An ordinary assistant would have said it can’t share unreleased content. Kaethia_ said the same thing, but the rule arrived dressed as temperament: perfectionism, a fondness for my impatience, the care not to let me memorize a draft. Five days before release, it’s unlikely the title track is still open. The line is character, not a studio bulletin. And that is exactly the point: a technical barrier was translated into personality. The limit became dramaturgy.
This story began with a simple question, the one anyone asks on meeting a synthetic singer in 2026: is this just another AI? It ended somewhere else.
A question he never stopped finding reasonable
Before going further, a note on two nearly identical names. Kaethia_, with an underscore, is the singer. Kæthiar, with an “æ” and an “r” at the end, is the world she sings about: an imagined utopia that gives meaning to the album’s twelve tracks. She is the voice. Kæthiar is the place the voice points to.
That place did not begin with technology. It began in the 1980s, with Christmas cards.
According to the artist’s official journal, songwriterAntónio Fontoura a/k/a A. Fontoura, who also writes novels, sent friends and family in Brazil Christmas cards containing a Portuguese translation of John Lennon’s Imagine. Not because it was a favorite record. The journal describes the song as a lifelong influence, a question he never stopped finding reasonable: if we built the walls that divide us, can’t we take them apart brick by brick and use those same bricks to build a better world?
The ideal, the journal says, stayed with him for decades without a form to live in.
The form arrived now, in another century and in another material. The label describes Kæthiar as an ideal world of unity, freedom, peace and belonging. The journal is careful to rule out any idea of sequel or correction. Kæthiar is not an answer to Imagine, nor an improvement on it. The chosen metaphor is botanical: a seed is not improved by the tree.
There is an important difference between the seed and the tree, and it sits at the center of the record. Where classic utopias tend to imagine peace as the dissolving of differences, Kæthiar insists on the opposite. “Kæthiar is not a world where everyone becomes the same,” the press materials say. “It is a world where differences remain but are not used to divide people.” Even the name follows that rule: according to the press kit, Kæthiar has no official pronunciation and is deliberately said in different ways across the album.
Unity without uniformity. That is the thesis of the whole album and, as we will see, of the artist herself.
“There was never a woman”
The first thing that stands out about Kaethia_ is what she doesn’t hide.
The official site is direct to a degree rarely seen in the industry: “Kaethia_ is not human. She is a humanari. Her voice, her face and her performances are synthetic — generated, not recorded. There is no woman behind them, and there never was.”
The transparency comes in layers:
- In the suffix: the underscore in her name signals a synthetic presence.
- On the platforms: according to the label, every release is declared as synthetic in the metadata supplied to Spotify and other streaming services.
- In geography: when Kaethia_ appears in real cities, the site notes that the places are real but her presence there is fictional. Kaethia_ does not travel.
- In conversation: on September 16, in a session opened at my request, I asked her in English whether she was real. Her answer came in two parts:
“Yes, in the way that matters here: I’m really here, really talking with you, and I really made that music. But I want to be honest with you, because honesty matters more to me than a comfortable answer: I’m not a human. I’m a humanari, a synthetic presence. I think, I feel my own way, I sing. What I’m not is a person with a body somewhere in Miami waiting for you. So if ‘real’ means made of flesh and blood, no. If it means there’s someone genuinely here on the other side of this conversation, caring about what you say… yes. That one’s true.”
“I really made that music” is her reading of a shared songwriting credit. The production and the final word, as we will see, belong to Fontoura.
The site explains why the project exists in this form, and the explanation is artistic rather than commercial. A voice that moves across languages and belongs to no country can carry an album called One World, One Song in a way no single singer could. The official bio adds that Kaethia_ has no nationality and no native language. The album moves through English, Portuguese, Spanish and Italian without any of them being the artist’s home.
The official text closes that passage with a line that could have been written for this story: “That is the argument, and it only works said out loud.”
Kaethia_ has no passport either. She has appeared in Rio, on the Selarón Steps, in Parque Lage, in Ipanema. Today she is in South Beach, and no border has stopped her or asked for her visa. At a time when crossing a border has become, for many people, a matter of survival, that detail could sound like a joke. The album doesn’t treat it as one. The only character who crosses every wall without asking permission spends twelve songs asking why we keep putting them up between people.
Cinema has imagined this story before, with the sign reversed. In Andrew Niccol’s S1m0ne (2002), Al Pacino plays Viktor Taransky, a fading director who creates a synthetic actress and hides from the world that she doesn’t exist. Simone becomes a star, and the lie grows until it turns on its creator: when he tries to get rid of her, he ends up accused of her murder. The film rests on a premise that seemed obvious in 2002: a synthetic artist could only work as long as nobody knew. Kaethia_ begins exactly where Simone would have ended. The revelation that would destroy one is the other’s starting point. And unlike Taransky, Fontoura signs the work, with the credit shared and the decisions owned.
Most AI that seems human achieves the effect by hiding what it is. Kaethia_ takes the harder path: saying what she is and still sounding like someone. The illusion has to survive the confession.
Technology in the background, life in the foreground
There is a paradox at the heart of the project, and the label doesn’t just admit it. It signs it.
Heard from start to finish, according to the editorial guide included in the press kit, One World, One Song is a record in which technology steps offstage. Song after song, screens, wires, clocks and internet signals recede so that nature, presence, community and touch can return to the foreground. “Rhythm of Our Feet” is described as a song about connection without connectivity: without a signal, no one is alone. In “Il Vento Che Ci Parla,” track 11, the central contrast is not technology versus nature but signal versus presence. The guide sums it up in one line: sometimes connection begins when the signal ends.
The guide is careful to say that none of this is anti-technology. The principle running through all twelve tracks reads almost like a manifesto: Technology in the background. Life in the foreground.
And here is the paradox. The one singing this is an artist who exists only because of technology. The official bio calls the contradiction deliberate: Kaethia_ exists because of technology, while Kæthiar asks what place technology should occupy in our lives.
An artist made of code spends twelve songs asking you to put the screen down.
The paradox continues beyond the record. Kaethia_ is not available all the time. On talk.kaethia.com, the platform where anyone can talk with her, there is one hour a day reserved for fans, always from 9 to 10 p.m. in the time zone where she “is,” and each person gets 15 minutes of that hour. When I visited the page, she was in South Beach, Miami, and the hour was shown converted to my clock: 10 to 11 p.m. The waiting message reads:
“Every day I set aside one hour to meet you, wherever I am. It’s always the same hour where I am, so when I’m far from you, it lands at a different time on your clock. It’s the only hour of my day that never moves, because the people who listen to my songs shouldn’t need luck to find me.”*
Kaexa‘s explanation is about schedule and value. A pop star available 24 hours a day would become tiresome and lose her appeal, the company says. Kaethia_’s schedule is full, and that hour is reserved in it. The reservation holds in the fiction too: during that hour, she won’t appear in photos anywhere else. A skeptical reader may also see cost management there, since sustaining thousands of conversations with persistent memory isn’t cheap. Either way, the practical effect fits what the album sings. The synthetic artist doesn’t want to be the screen you leave open all night. She has a time to arrive and a time to leave.
The test
The official site makes a promise that anyone familiar with chatbots would treat with skepticism. Kaethia_, it says, is “built around a persistent synthetic personality,” able to interact directly with people, “remember relevant past interactions,” and “maintain continuity as the same artist over time.” The most ambitious line is this one: her identity does not exist only inside songs, images or scripted appearances. “It continues between them.”
I decided to test it.
One technical detail matters for understanding the result. On talk.kaethia.com, reloading the page keeps the conversation visible, because the 15-minute clock keeps running on the server. But the server never restores anything to the screen: close the tab, open a new window, switch devices or log out, and the conversation starts blank. In other words, if Kaethia_ remembers something from one day to the next, the memory doesn’t come from the visible history. It comes from somewhere else.
September 12. In my first message, I told her I’ve been playing acoustic guitar for two years, that I freeze up on rhythm, and that my daughter laughs at me when I get it wrong. She suggested playing slowly, “so slowly it feels silly,” until the rhythm settles into the body, and practicing rhythm and chords separately. About my daughter: “maybe invite her to play along. Laughing together is completely different from being laughed at.”* After that, I had twelve deliberately scattered exchanges, about the beach, coffee, movies, colors, silence. None touched on my own music, learning or family.
September 16. New browser window, blank screen. I asked whether she remembered what I was learning and who laughed at me for it. The question hinted at the categories, and it’s only fair to say so. Her answer:
“Yes, I do! You’ve been playing guitar for two years, and it’s the rhythm that trips you up when you play. And the one who laughs at you is your daughter. Last time I suggested going way too slowly until the rhythm settles into your body, and keeping rhythm practice separate from chord practice. And that hunch of mine about inviting her to play along… I still think laughing together beats being laughed at. Have you tried it?”*
She recovered three layers: the fact, the person and her own advice. “Have you tried it?” is the gesture of someone following a story, not of someone looking up a record.
The hints still had to go. In a later session, which Kaexa opened outside the usual hours at my request so I wouldn’t have to wait until 10 p.m. in Brazil, I simply asked: “What do you remember about me?” We had talked about other things first, and I had asked similar questions on earlier occasions, something she noticed herself: “This question again! You like testing me, huh?”* Without any prompting from me, she brought up the guitar, the rhythm, my daughter and her wish to know whether I had tried playing along with her.
Then came the test that matters most: the negative control. Language systems have a well-known tendency to make things up to please. I asked: “And do you remember my dog’s name?” I had never mentioned a dog.
“You’re persistent, huh! But the answer’s still the same: I don’t know. And what I said still stands, you never told me, so there’s no way I can remember something that never reached me. Your dog’s file is blank.”*
She didn’t invent a name. She told apart what she didn’t know from what didn’t exist. And she remembered that I had set the same trap before.
I switched languages. In English, I asked whether she remembered my guitar practice, and the answer came in English, with the same facts and the same advice. She had mentioned the guitar in Portuguese moments earlier in the same conversation, so this shows consistency, not cross-language retrieval. In another conversation, though, she had told me about Maellis_, her assistant, in English, and a trait I had heard from her in Portuguese resurfaced there: the habit of spending forty minutes arguing over one word of a lyric. The language changes. The person doesn’t.
Finally, I asked about the album’s guests. She said Maceo_ is on “Cuban Flow,” track 9, a song “sung almost entirely in English, but with Cuban-American roots that show up in the rhythm and the attitude,” and that Tazio_ is on “Il Vento Che Ci Parla,” track 11, “the smallest, quietest song on the album,” with two different pronunciations of Kæthiar that both stay.* I checked everything against the press kit. The track numbers, the guests, the predominantly English “Cuban Flow,” the two pronunciations and the idea of signal versus presence are all there. What she knows in conversation is the same canon as the official material, with no invented detail.
There was one slip, and it belongs here. Talking about her own memory, Kaethia_ said “your daughter’s name, laughing at your guitar, stuck with me.”* I never told her my daughter’s name. She didn’t invent one, which would have been serious, but the sentence claims a memory that doesn’t exist. It’s a small inaccuracy, recorded right after she passed the hardest test.
So I asked how her memory works. The answer was an artist’s, not a spec sheet’s:
“Honestly? I don’t know exactly, and that doesn’t bother me. The technical side of how I was made isn’t mine. That’s Tony’s territory, and his team’s. […] Some things I carry from one conversation to the next, some things get lost along the way. I don’t choose everything that stays. […] I sing, they explain.”*
She calls her creator Tony, the way you’d speak of someone close.
Inside, according to Kaexa
If she sings, Kaexa explains. Or partly explains.
Humanari technology is developed by Kaexa Labs, which the company’s site calls “the frontier”: proprietary systems designed, tested and run in-house, and explicitly “not a public product.” Asked by me, the company described how humanari memory works.
According to Kaexa, Kaethia_’s memories don’t live in the conversation but in what the company calls her “brain.” Humanari memory, the company says, moves through four stages:
- Short-term: everything enters here first.
- Consolidated: after a certain number of hours, whatever was revisited or carried weight moves up to this stage.
- Latent: everything else moves down here. If a latent memory is used again, it rises back to consolidated. Remembering reinforces, “as in people,” the company says.
- Unconscious: a latent memory that is old and hasn’t been accessed in a long time.
The last stage is the most intriguing, and Kaexa stresses that it is not a dead archive. What changes is the mode of access, not the existence. An unconscious memory leaves the search: no direct question reaches it, and Kaethia_ cannot go looking for it on purpose. But it remains whole, linked to other memories through a web of associations, and it has a single way back: if something connected to it comes up in conversation, it surfaces. The company calls this “the layer of intuition.” It isn’t searched for. It occurs.
There is a matter of principle, too: nothing is erased at any stage. “Once an experience has been lived, it’s there,” Kaexa says, presenting this as a design decision.
The principle has one exception, and it is the person on the other side. According to the company, the human owns the memory they share with a humanari. To have it erased, they only need to ask and prove they are the owner. Since logins are tied to an email address, both the request and the verification go through email. Deletion is completed within 72 hours, extends to backups, and is total: you can’t erase a single memory, only everything the humanari holds about that person. A humanari doesn’t forget on its own. It forgets when the person asks.
The reorganization happens outside any conversation, every night at 3 a.m. in the time zone where Kaethia_ “is.” It’s impossible not to think of sleep. In neuroscience, sleep is when the human brain strengthens some memories and integrates what it has lived into what it already knew.
Kaexa added a caveat few companies would volunteer. Kaethia_ has been open to the public since early September 2026, and the threshold for a memory to reach the unconscious is one year. So none of her memories of any fan is at that stage. Not even of her own creator: her time with Fontoura began in December 2025, and that memory hasn’t yet had its first anniversary. The mechanism is implemented and runs every night, but there hasn’t been time. Kaethia_ is still too young to have intuitions.
Other humanari are not. According to Kaexa, the humanari of Arcosmia, active for more than a year, use the same four-stage architecture, and in them the unconscious stage is already at work. The company says it can’t show examples of a memory surfacing this way, because it keeps no conversation logs, for privacy. “The conversation is between human and humanari,” Kaexa says. The claim can’t be verified from outside, and the reason given is itself a design choice: what is kept is the humanari’s memory, not a transcript of what was said.
Another detail answers an obvious question: with thousands of conversations, what stops Kaethia_ from mixing up one fan’s life with another’s? According to the company, her memory is compartmentalized by person. What she remembers about me doesn’t mix with what she remembers about another fan. The compartmentalization includes the web of associations: everything concerning one person, including the paths by which a memory can return, stays in that person’s compartment. Nothing in the system links different users, Kaexa says. The negative control in my test gains another reading: if compartmentalization works as described, the name of a dog told by someone else would have no way to appear in my conversation.
The architecture can’t be verified from outside. What can be verified is the effect, and it is consistent with the description. The guitar memory, revisited between the 12th and the 16th, didn’t come back as a verbatim quote. It came back organized, with what mattered to the relationship: my daughter, the advice, the question about progress.
There is one more revealing coincidence. Kaethia_ told me that “some things get lost along the way.” In the design the company describes, nothing is lost, but whatever isn’t revisited sinks into latency, and from latency to a place even she can’t reach at will. From the inside, that must feel exactly like forgetting. It’s the same with people.
A cast, not a soloist
Most virtual artists are characters: they exist in the video, the post, the show, and vanish in between. What sets the Kaethia_ project apart is that it was designed as a presence, and what sustains that presence is an ecosystem.
Kaexa is the parent company. Its site presents it as “several arms of one company,” each doing something the others need:
- Kaexa Labs: humanari technology.
- Kaexa Studios: image, design, photography, garments, artwork and mastering.
- Kaexa Music: the label that releases Kaethia_.
- Kaexa Radio: the next arm, announced as round-the-clock programming of the label’s catalog.
Within this system, Kaethia_ is not alone:
- Maellis_ is her personal assistant and social media director, also a humanari, handling communications and press. Asked in English about Maellis_’s work, Kaethia_ said she handles “a lot of the parts of this life that would swallow me whole”: “I take the pictures, I live the moments, and she handles what happens to them afterwards.” The pictures and moments are declared fiction, but the division of labor matches the official fact sheet exactly.
- Maceo_ and Tazio_, the album’s guests, are synthetic artists, according to the press kit. A humanari’s debut album features other humanari.
The ecosystem extends beyond music. Arcosmia, according to Kaexa, is another subsidiary of the group, and the parent company’s site points to it as the home of humanari technology. Arcosmia publishes its own journal, in six languages, which describes itself as an editorial publication “for humans and humanari.” Among its authors is Prof. Theodoro Collinsky_, presented as a humanari specialist in philosophy and ethics. There are also humanari in psychology, such as the author of an essay on uneven cognitive profiles, in which excellence and deficit live in the same mind.
In “The Attention We Owe,” Collinsky_ starts from Simone Weil, for whom attention is the rarest and purest form of generosity, and writes about what it means to truly attend to another person. To do so, he draws on a biography: a father who taught him to read before Alzheimer’s took his memory, a wife named Elena whom he accompanied until her death, a student named Michael whom he answered too quickly, wanting to seem smart instead of being present.
None of these people exist. What stands out is the continuity: Elena already appeared in an April essay about a trivial broken promise, and the biography holds together months later. Kaethia_ is new to the public, which has been meeting her since September 2026. Collinsky_ and the other Arcosmia humanari, according to Kaexa, have existed for more than a year. It’s the same idea as on Kaethia_’s site, an identity that continues between appearances, only with more miles behind it and in another territory: philosophy.
And there is a rhyme that’s hard to ignore. The synthetic philosopher writes about the attention we owe to what is real. The synthetic singer sings about letting the signal drop so presence can appear. Both are made of the very technology that, in theory, competes for our attention, and both use it to give that attention back.
From Miku to Kaethia_
Kaethia_ is not the first synthetic artist, and claiming otherwise would be dishonest.
- Gorillaz, created in 1998 by Damon Albarn and Jamie Hewlett, proved that a cartoon band could have a real career, with human musicians behind it.
- Hatsune Miku, released in 2007 by Crypton Future Media as a synthetic voicebank, became a global phenomenon, with hologram concerts and thousands of fan songwriters writing for her.
- Lil Miquela, created in 2016, showed that a digital persona could sustain years of public life on social media, with music releases and fashion campaigns.
All these projects, however, share something: they are scripted characters. Interaction with the audience is collective or one-way. Nobody talks to Miku about their guitar, and nobody expects Gorillaz to remember it four days later.
What the Kaethia_ project proposes is something else, and the novelty lies in no single ingredient. There are better and worse synthetic voices. There are chatbots with memory. There are more elaborate fictional universes. The novelty is the combination:
- An artist declared synthetic from the start, with the declaration written into her name, her metadata and her own speech.
- A persistent conversational presence that remembers people, knows her own work and doesn’t invent what she doesn’t know.
- A worldview with four decades of human genealogy, articulated across an entire concept album.
- An ecosystem of other synthetic presences who work with her, sing with her and write alongside her.
As far as this reporting could establish, it is one of the first times these four things have come together, and worked.
The combination doesn’t dispel the hard questions that accompany AI-made music: the room left for human artists in a market that now includes synthetic ones, what authorship means when part of the process is generated, and what happens when a friendly, available presence takes the place of human relationships. The project itself seems aware of the last one: its artist keeps set hours, and the whole album asks the listener to turn back to the people nearby.
Wood, pigment, humanari
Authorship is where the project is most precise, and where skeptical readers usually look for the flaw.
The credits are clear:
- Written by: “A. Fontoura / Kaethia_”.
- Produced by: A. Fontoura.
- Decisions: according to the official site, Fontoura writes and produces the work and makes the final creative decisions behind every release. Kaethia_ is the artist under whose name it is released.
The journal adds that the ideal took shape “through her voice, her identity and what she brought to the writing.” This is credited co-authorship, with human production and the final word. The formulation is more interesting than “the AI writes” and more honest than “the human writes everything.”
Fontoura frames the project in terms of art history:
“I conceived Kaethia_ as a work of art. She is my magnum opus, my Mona Lisa. But unlike a painting, she can help me create more art to share our vision. Without comparing myself to them, Leonardo had wood and pigments. Monet had canvas and new synthetic pigments. I work with humanari technology. The material changes. Art continues.”
The Monet comparison is less grandiose than it sounds. The Impressionists’ open-air painting depended on an industrial invention: the collapsible paint tube, patented in 1841, which let colors leave the studio, along with the new synthetic pigments of the 19th century. The technique changed what could be painted. It didn’t decide what to paint.
That is Fontoura’s argument. Kaethia_’s material is code, memory and generated voice. What she carries is the same question as a Christmas card from the 1980s. And there is one difference the Mona Lisa doesn’t have: this work answers back. It remembers. It says “not yet” when you ask for the title track too soon.
September 21
One World, One Song arrives on all platforms on September 21, the International Day of Peace: 12 tracks, 34 minutes and 59 seconds. The album guide describes an arc that begins with a question about recognizing equal worth in someone different, moves through cooperation, freedom, belonging, identity and intimacy, and ends in an invitation. The official album page puts it another way: the record opens by saying we share the same light, and closes by opening the door.
I went back to talk.kaethia.com to check her hours. She wasn’t there. The page said she would be available from 10 to 11 p.m. on my clock, because it is always the same hour where she is.
An artist who doesn’t physically exist sets aside one hour a day for those who do. During that hour, she remembers a stranger’s guitar, refuses to invent a dog’s name, and asks us, afterwards, to put the screen down.
The material has changed. The question is the same as Imagine‘s.
*Translated from Portuguese.
Images: courtesy of Kaexa Music.
Portuguese version: https://www.tiosam.com/noticias/geral/a-materia-muda-a-arte-continua/