Carole & Tuesday Saw It Coming: What AI Music Can Copy in 2026, and What It Can't

In 2019 an anime imagined a Mars where machines write the hits and almost nobody minds. Two teenagers, a refugee orphan with a keyboard and a runaway with a guitar, start writing their own songs anyway, and the story treats that as a quiet act of rebellion. Seven years later, back on Earth, one streaming service receives nearly 90,000 fully AI-generated tracks every day. Carole & Tuesday did not predict the technology. It predicted the argument.

The show came from Studio Bones and supervising director Shinichirō Watanabe, the director of Cowboy Bebop, and it aired before ChatGPT, Suno or Udio had reached the public. Its Mars is a place where artificial intelligence produces most culture and people mostly consume it. The machine-made pop in the show is polished, popular and slightly airless. What it lacks is harder to name, and the series spends 24 episodes trying to name it.

My argument is this: the show was right that AI would make competent, chart-ready music at industrial scale, and right about what that leaves scarce: a life behind the song, a stage, trust, and someone who answers for the work. The fights of 2026 over training consent, disclosure, voice cloning, royalty dilution and discovery are the ones the show kept in the background, and they will decide whether that scarcity is rewarded or exploited.

What follows recaps what the show imagined (spoiler-light), offers a simple model of what AI can and cannot copy, and tests it against the 2025–26 evidence on uploads, listeners, live music and the law, including two viral Filipino songs. The show’s back half comes later, behind a spoiler warning.

Play · YouTube

Context: why this matters now

When an interviewer from Looper asked Watanabe about AI art programs, which had arrived in the news after the show was made, he answered: “The future that I imagined has arrived.” He added that the images such programs made were often strange and distorted, so he thought the full vision was still a long way off. Music has moved faster than his caution.

The numbers now have the shape of the show’s premise. Suno, the best-known AI song generator, was valued at $5.4 billion after raising more than $400 million in June 2026, and Deezer says AI tracks are more than half of its deliveries on peak days.

Exhibit 1

90Kfully AI-generated tracks reach Deezer each day (nearly)DEEZER · OCT 2026

97%of 9,000 people in a blind test could not identify the AI tracksDEEZER / IPSOS · NOV 2025

100M+people who have used Suno, according to its CEOFORBES AUSTRALIA · FEB 2026

$5.4BSuno valuation after its June 2026 funding roundSUNO · JUN 2026

Deezer figures cover one platform and rely on its own detection tool. Suno's user figure is the company's own claim.

Source: Deezer (13 Oct 2026); Deezer/Ipsos (12 Nov 2025); Suno (3 Jun 2026); Forbes Australia (27 Feb 2026)

The response has been just as fast. In the space of about two years, AI music went from a curiosity to a matter for courts, legislatures, chart compilers and streaming policy teams. The timeline below is the short version.

Exhibit 2

1. APR 2019Carole & Tuesday premieresOn Fuji TV’s +Ultra block in Japan; Netflix streams it worldwide from August 2019.
2. MAR 2024Tennessee signs the ELVIS ActVoice joins name and likeness as a protected right; in force from 1 July 2024.
3. JUN 2024Major labels sue Suno and UdioUniversal, Sony and Warner allege training on their recordings without a licence.
4. JAN 2025US Copyright Office: prompts are not authorshipPurely AI-generated material cannot be copyrighted; AI-assisted work can be.
5. OCT–NOV 2025The first settlementsUniversal licenses Udio; Warner licenses Udio, then Suno.
6. JAN 2026Bandcamp bans AI-made musicMusic made wholly or substantially by AI is no longer allowed.
7. MAR 2026Supreme Court declines ThalerThe human-authorship rule stands.
8. JUL 2026Labels for AI music; GEMA beats SunoAn IFPI-led coalition defines AI-generated and AI-assisted; a Munich court rules against Suno.
9. AUG 2026Spotify and ARIA draw linesSpotify badges AI Persona profiles; Australia’s charts exclude wholly AI tracks.
10. SEP 2026Suno ships licensed v6; Universal and Sony sue againA new complaint over 60,202 recordings, filed 18 September.

Source: Wikipedia; AP; IPWatchdog; TechCrunch; Music Week; Bandcamp; Supreme Court of the United States; IFPI; Digital Music News; ARIA; The Next Web

The show: a Mars where machines write the hits

This section covers the setup and the first half of the series only.

The official synopsis sets the story about fifty years after humanity began moving to Mars, in an era when AI produces most culture and people are mostly its consumers. Carole is a part-time worker in Alba City who wants to be a musician. Tuesday comes from a wealthy family in Herschel City and wants the same thing, with nobody at home who understands it. They meet by chance on a bridge and decide to make music together.

The two are drawn as opposites. Anime News Network’s reviewer described Carole as an orphaned refugee from Earth who busks with a keyboard between part-time jobs, and Tuesday as a politician’s daughter who runs from her privileged life with an acoustic Gibson guitar. Around them gather a washed-up former rocker, Gus, who appoints himself their manager, and a young sound engineer, Roddy, whose video of them goes viral.

The machine side of the world arrives in the first episode. Angela, a former child actress and model, is taken on as a singer by Tao, a producer who uses advanced AI to build high-selling performers. The ANN review describes Tao using AI and mechanical enhancement to shape her, so that algorithms decide every syllable she sings. A star DJ, Ertegun, scoffs at the very idea of music not made by AI. A cheap robot video director the duo hire turns out to be a scam. And the first half builds towards Mars’ Brightest, a televised talent show where the duo’s self-written songs and Angela’s AI-shaped act compete for the same prize.

Two details make the show a good lens for this argument. First, its episodes are named after real songs: “True Colors”, “Born to Run”, “Fire and Rain”, “Video Killed the Radio Star”, “Every Breath You Take”, “Dancing Queen”, and a finale called “A Change Is Gonna Come”. The titles quietly insist that music is a lineage of human work. Second, the show practised what it preached. Its score is by the composer Mocky, its contributors include Flying Lotus and Thundercat, and its leads’ singing voices belong to the singers Nai Br.XX (Carole) and Celeina Ann (Tuesday). Mocky won Best Score at the Crunchyroll Anime Awards. A story about human music was made by a roster of named humans.

ANN’s reviewer called the show’s future “a logical endpoint” of how present-day society already values music. That is the right frame. The show is not science fiction about robots. It is a satire of an industry that already optimised for volume, polish and data, extended fifty years.

A framework: what AI can copy, and what it can’t

To test the show against today’s evidence, I use a five-part model. A piece of music reaches a listener through five layers, and AI’s reach is very different at each one.

• Sound is the audio itself: timbre, mix, polish, the feel of a genre.
• Song is the composition: melody, harmony, structure, lyrics.
• Story is the life the listener believes is behind the song: who made it, why, what it cost them.
• Stage is presence: a performance in a shared room, at a set time, that cannot be repeated exactly.
• Stake is accountability: someone who consented, gets paid, can be credited, and can be held responsible.

Exhibit 3

	Can AI copy it?	Evidence, 2025–26	What stays scarce
Sound	Yes, convincingly	97% of 9,000 people could not pick out AI tracks in a blind test	Very little: polish is now cheap
Song	Largely, at a competent level	At least six AI or AI-assisted acts entered Billboard charts in a few months of 2025	Distinctive writing, often still supplied by a human
Story	It can fake one, not live one	Listeners rate identical music lower when told it is AI; invented backstories unravel	A credible life behind the song
Stage	No	Live Nation drew a record 159 million fans in 2025	Presence in a shared room
Stake	No	US law requires a human author; German courts hold AI firms responsible for outputs	Someone who consents, is paid and answers for it

Framework by the author. Evidence is drawn from the sources cited in this article; see the deep dives for detail. The table scrolls sideways on small screens.

The model makes one prediction that the rest of this article tests. If AI makes sound and song abundant, their price falls towards zero, and value moves to the layers AI cannot supply. That means more weight on disclosure (so listeners can find the story), on live performance (the stage), and on consent and liability (the stake). It also predicts where the fights will be: wherever someone tries to borrow the scarce layers without paying for them, by faking a story, cloning a voice or training on a catalogue without permission.

Deep dive 1: sound and song, where the flood is real but the audience is not

Deezer publishes regular data on fully AI-generated uploads, because it built its own detection tool in January 2025 and began tagging AI music in June 2025. Its series is the clearest public picture of the supply side.

Exhibit 4

10K

30K

50K

60K

75K

90K

Jan 2025Sep 2025Nov 2025Jan 2026Apr 2026Oct 2026

Approximate daily figures at each announcement; intervals are uneven. Share of all daily deliveries: about 34% in November 2025, 44% in April 2026 and more than 50% on peak days by October 2026.

Source: TechCrunch, 20 Apr 2026 (January 2025 to April 2026); Deezer, 13 Oct 2026 (October 2026)

View data table

Announcement	AI tracks per day	Share of daily deliveries
January 2025	about 10,000	not reported here
September 2025	about 30,000	not reported here
November 2025	about 50,000	over 34%
January 2026	about 60,000	not reported here
April 2026	almost 75,000	44%
October 2026	nearly 90,000	over 50% on peak days

Where does that volume come from? Mostly from tools that turn a text prompt into a finished song in seconds. Suno is the market leader. A Billboard report on its investor materials, cited by Forbes Australia, put its output at about 7 million songs a day. Its chief executive said in February 2026 that it had passed 2 million paid subscribers and $300 million in annual revenue, up from the $200 million it reported when it raised $250 million at a $2.45 billion valuation in November 2025. Seven months later it was worth more than twice that.

Here the show’s prediction is both right and wrong. Right: AI really can make competent music at a scale no human industry could match. Wrong about the shape: Carole & Tuesday imagined AI music as a top-down studio product, Tao’s lab manufacturing a handful of polished stars. The real thing is a consumer app used by tens of millions of people, producing an ocean of tracks that almost nobody hears.

Exhibit 5

Share of uploads, peak days>50%

Share of all streams1–3%

AI-track streams that were fraudup to 85%

The three bars use different bases: all daily deliveries on peak days; all streams on the platform; streams on fully AI-generated tracks in 2025. Fraudulent streams are excluded from royalty payments.

Source: Deezer, 13 Oct 2026

View data table

Measure	Share
AI share of daily deliveries, peak days (2026)	more than 50%
AI share of all streams	1–3%
Share of streams on AI tracks found to be fraudulent (2025)	up to 85%

This gap is the most important number in the debate. Fully AI-generated music is a large share of what is uploaded and a tiny share of what is chosen. And up to 85% of the streams those tracks did attract in 2025 were fraudulent, according to Deezer, which strips such streams out of royalty payments. Much of the flood is not an artistic movement. It is an attempt to skim the royalty pool with cheap content and fake plays.

That is why the economics matter even if listeners ignore AI tracks. Streaming royalties are paid from shared pools, so every fraudulent stream that is not caught dilutes the payment to everyone else. Deezer cites a study by the collecting-society body CISAC with PMP Strategy estimating that nearly a quarter of creators’ revenues are at risk by 2028.

The platforms’ first response has been hygiene rather than philosophy. Deezer removes detected AI tracks from algorithmic recommendations and editorial playlists, no longer stores hi-res versions of them, and says it detected more than 42 million AI-generated tracks and kept them out of recommendations between January 2025 and August 2026. In October 2026 it announced a setting, due before the end of the year, that lets listeners hide fully AI-generated tracks, albums and artists from search. Spotify said in September 2025 that it had removed more than 75 million spam tracks in the previous twelve months and announced a spam filter aimed at mass uploads, duplicates and artificially short tracks.

What this means for the framework: sound and song are no longer scarce, and the cost of producing them is close to zero. But abundance has not translated into attention. The scarce resource on Mars was never the hit; it was the listener who cared.

Deep dive 2: story, and what changes when listeners know

If sound is copyable, the next question is whether listeners care where music comes from. The evidence says: not when they cannot tell, and quite a lot when they can.

Exhibit 6

Could not spot the AI tracks97%

Want AI music labelled80%

Want to know if recommended73%

Would try it out of curiosity66%

Keep it off the main charts52%

Want to filter it out45%

Online survey of adults aged 18–65 in the US, Canada, Brazil, the UK, France, the Netherlands, Germany and Japan. The recommendation, curiosity and filter questions were asked of music-streaming users. Survey commissioned by Deezer, which sells AI detection.

Source: Deezer/Ipsos, 12 Nov 2025; Deezer, 13 Oct 2026 (filter figure)

View data table

Statement	% agreeing
Could not correctly identify the fully AI-generated tracks	97%
100% AI-generated music should be clearly labelled	80%
Want to know if a service recommends 100% AI music	73%
Would listen to 100% AI music at least once out of curiosity	66%
100% AI songs should not appear in main charts	52%
Want to filter out 100% AI music	45%

Two findings sit side by side. Ninety-seven per cent of respondents failed the blind test: they heard two AI tracks and one human track and could not identify which was which. But 80% want fully AI-generated music clearly labelled, 52% think it should stay off the main charts, and 45% of streaming users would like to filter it out. People are curious about AI music (66% would try it once). They also want to know what they are listening to.

Laboratory studies explain why the label matters. In a 2023 paper, researchers at Missouri University of Science and Technology found that listeners liked music less when they believed an AI had composed it. The effect was conditional: when they told participants that electronic music was AI-made, liking did not change, which the authors put down to electronic music already sounding machine-like. With human-sounding classical excerpts, the AI label lowered liking. A 2025 study in Empirical Studies of the Arts went further. One hundred and twenty participants watched piano performances with identical audio, once attributed to a pianist and once to an AI-driven piano. They rated the “human” versions as more likeable, more engaging and of higher quality, and when asked what differed, they described differences in rhythm, tempo and dynamics that were not there. Belief about who made the music changed what people heard.

That is the story layer in action. A listener’s experience of a song includes their model of the person behind it. Take the person away, or reveal that there never was one, and the same audio lands differently for many listeners. The show dramatised exactly this: Angela’s songs are flawless, but the series keeps handing its emotional payoffs to the duo’s rougher, self-written songs, which are plainly theirs.

The industry is now building disclosure into its plumbing:

• Labels. On 10 July 2026 an IFPI-led coalition including the RIAA, the Recording Academy, SAG-AFTRA and the Human Artistry Campaign launched voluntary track-level labels. “AI-Generated” means generative AI produced all or the primary part of the creative elements, such as the lead vocal. “AI-Assisted” means humans performed the lead vocal and primary instruments and AI supplied some elements.
• Credits. Spotify backed an industry standard for AI disclosures in song credits, and in April 2026 launched a beta letting artists say how they used AI. It is candid about the limit: “the absence of a credit doesn’t mean AI wasn’t used.”
• Identity. In August 2026 Spotify began labelling “AI Persona” profiles, artist pages whose identity is AI-generated, and excluding their music from editorial and algorithmic recommendations by default unless a listener follows them. The label is about whether the artist is a real person, not about how the audio was made.
• Charts. From the chart dated 31 August 2026, Australia’s ARIA charts exclude wholly AI-generated recordings; AI-assisted ones remain eligible if they are substantially human-made and raise no manipulation concerns. Ineligible tracks also cannot win ARIA Awards.
• Storefronts. Bandcamp, a marketplace where fans buy directly from artists, banned music generated wholly or in substantial part by AI in January 2026, along with using AI to imitate other artists.

Notice where every one of these lines is drawn: not at “any AI use”, but at the human role. That is the framework’s story and stake layers turned into policy.

Deep dive 3: stage, and what two people in a room add

The emotional core of Carole & Tuesday is not a protest against machines. It is a partnership. Two people who could not be more different discover that their halves of a song fit together, and the show keeps returning to the moment when they play for someone: on a bridge, in an empty concert hall, in a bar for an audience of ten, on a festival stage in front of a hostile crowd.

Play · YouTube

The stage is the layer AI cannot reach at all. A generated track can be streamed a billion times. It cannot stand in a room with you at a particular time, make a mistake, recover, and look at the person next to it. If the framework is right, the value that drains out of recorded sound should show up in live presence. The 2025 data is consistent with that.

Exhibit 7

159Mfans at Live Nation events in 2025, up 5%LIVE NATION · FEB 2026

$25.2BLive Nation revenue in 2025, up 9%LIVE NATION · FEB 2026

55Kshows worldwide in 2025 (approximately)LIVE NATION · FEB 2026

67Mfans with 2026 concert tickets by early February (approximately)LIVE NATION · FEB 2026

One company's results, reported by the company. Live Nation is the world's largest promoter, so its figures reflect large-scale touring more than grassroots venues.

Source: Live Nation Entertainment, 19 Feb 2026

Live Nation reported record revenue of $25.2 billion and a record 159 million fans in 2025, with international markets overtaking the US in attendance for the first time. By early February 2026 about 67 million fans had already bought tickets for its 2026 concerts, a double-digit increase on the same point a year earlier.

This is not proof that AI caused the live boom; the trend began long before AI music. Nor does it mean every musician benefits: big promoters’ numbers are dominated by stadium acts, and small venues face different economics. But the direction matters for the argument. The parts of music that require a body in a room are growing, at exactly the moment the parts that can be generated are flooding. Suno itself seems to agree: as part of its Warner deal in November 2025, it acquired the live-music discovery service Songkick.

What this means for the framework: for a working artist, the recording is increasingly the advertisement and the room is the product. The show got this right. Carole and Tuesday never win on production values; their breakthroughs come when they play for people in a room.

Deep dive 4: stake, meaning consent, credit and who answers for the work

The last layer is the one the show treated most lightly and the real world is fighting over hardest. Stake covers three questions: did the people whose work trained the model consent, who owns and is credited for the output, and who is liable when it goes wrong.

Consent and training. In June 2024 Universal, Sony and Warner sued Suno and Udio, alleging that both had trained on their recordings without permission. Two years on, the fronts look like this:

Exhibit 8

	Suno	Udio
Universal Music	Still suing; filed a second suit on 18 Sep 2026 over 60,202 recordings	Settled with a licensing deal, Oct 2025
Warner Music	Settled and licensed, Nov 2025; named partner on v6	Licensing agreement, Nov 2025
Sony Music	Still suing; co-plaintiff on the second suit, Sep 2026	Still suing; filed a second suit in Jul 2026 over 30,000 more songs
Other licensors	BMG (Aug 2026) and Believe (Sep 2026)	Not covered here

Settlement terms were not disclosed. Lawsuits are allegations, not findings. The table scrolls sideways on small screens.

Source: Music Business Worldwide (29 Oct and 19 Nov 2025); Music Week (26 Nov 2025); The Next Web (18 Sep 2026)

The settlements changed the products. Under its Warner deal, Suno said it would launch licensed models in 2026, restrict downloads to paying users, and let artists and songwriters opt in to the use of their name, image, likeness and voice. In September 2026 it launched v6, a model family it says was built with Warner, BMG and Believe. Universal and Sony responded on 18 September with a second suit in federal court in Massachusetts, after the judge in the original case refused in August to let them add more than 61,000 works to it. They argue that Suno’s own licensing deals prove a market for training licences exists. Suno’s product chief has said the revenue-share deals are not in exchange for training.

Europe is moving faster through the courts. In November 2025 the Munich Regional Court ruled for GEMA, the German music rights society, against OpenAI over the lyrics of nine songs, including Herbert Grönemeyer’s “Männer”. It found that the models had memorised the lyrics and could reproduce them, rejected OpenAI’s reliance on the EU’s text-and-data-mining exceptions, and held OpenAI rather than its users responsible. On 31 July 2026 the same court ruled for GEMA against Suno, finding it liable for soundalike outputs and for models stored on German servers, and rejecting a US fair-use defence for the training copies because the works resurfaced in the outputs. Both rulings can be appealed, and Suno has said it disagrees.

Credit and ownership. In the US, purely machine-made music cannot be owned. The US Copyright Office concluded in January 2025 that prompts alone do not give a user enough control to be an author, while AI-assisted works can still be protected for their human contributions. The courts agree: the D.C. Circuit upheld the human-authorship requirement in Thaler v. Perlmutter in March 2025, and the Supreme Court declined to hear the case on 2 March 2026. A fully generated song can top a chart, but in the US no one owns it.

Voices and likeness. Tennessee’s ELVIS Act, signed on 21 March 2024 and in force from 1 July that year, added a person’s voice, including simulations of it, to the likeness rights the state already protected. At federal level, the NO FAKES Act, which would create a right over digital replicas of a person’s voice and likeness lasting up to 70 years after death, cleared the Senate Judiciary Committee unanimously on 18 June 2026 and awaits a full Senate vote. Spotify’s policy since September 2025 allows vocal impersonation only when the impersonated artist has authorised it.

Consent inside the industry. The newest front is between artists and their own partners. On 22 June 2026 a coalition of artist, songwriter and manager organisations coordinated by the European Music Managers Alliance warned labels and publishers that creators were receiving letters saying their work had already been placed into AI licensing deals without discussion. They asked for three things: the right to refuse, a clear split of revenue, and information before any deal is signed.

Case examples

The Velvet Sundown: a story that did not survive disclosure. In mid-2025 a “band” called The Velvet Sundown released about 40 tracks in roughly a month and passed a million monthly listeners on Spotify, helped by playlist placements. Amid speculation about who was behind it, a post from an account presenting itself as the project’s official voice described it as a synthetic music project guided by human creative direction. The music had not changed; the story had, and the conversation moved from the songs to whether listeners had been misled. The lesson: AI can supply the sound and the song, but a fabricated story is a liability that grows with success.

Xania Monet: AI sound, human stake. Xania Monet is an AI-voiced R&B act created by Telisha “Nikki” Jones, a Mississippi songwriter who, according to Billboard, writes about 90% of the lyrics herself and uses Suno to produce the songs. Monet topped Billboard’s R&B Digital Song Sales chart and reached No. 3 on Hot Gospel Songs, and Hallwood Media signed her to what Billboard described as a multimillion-dollar deal. Billboard estimated that five of her songs earned about $52,000 in just over two months. The lesson: the commercially durable AI acts so far have a real person with something at stake behind them. That is the framework’s stake layer, supplied by a human.

Breaking Rust: a song that charts without a singer. Billboard counted at least six AI or AI-assisted acts that entered its charts in a few months of 2025, including the AI country act Breaking Rust, which appeared on the Emerging Artists and Country Digital Song Sales charts. The lesson: the song layer is genuinely copyable at a level that sells. But download charts are small markets, which is why chart compilers are now deciding what they are meant to measure.

Bandcamp: a marketplace that sells the story. Bandcamp’s January 2026 policy bans music made wholly or substantially by AI and AI impersonation of other artists, and says it may remove music on suspicion of being AI-generated, so that fans can trust that what they buy was made by people. The lesson: for a direct-to-fan business, “made by a human” is not a moral stance. It is the product.

Closer to home: two viral Filipino songs and the question of who made them

The Philippines produced 2026’s clearest local test of this argument. Filipino listeners are loyal to local acts: in Spotify’s 2025 Wrapped for the Philippines, eight of the ten most-streamed songs were local, led by Cup of Joe’s “Multo”. That loyalty is to people, so the country’s AI-music arguments have been about whether there were people behind the songs at all.

There was a warning shot first. In October 2025 a rock cover of Cup of Joe’s “Pahina” made with AI tools and released by an act called Renegade Stories briefly reached No. 1 on Spotify Philippines’ Viral Songs chart. Band members reacted with open disbelief online, fans called it unfair to the original artists, the cover’s makers said AI had been used only as a tool to enhance the recording, and the cover was later removed from Spotify.

“Hawak Mo ang Beat”: the accusation became the story. “Hawak Mo ang Beat” (“You Hold the Beat”) is a dance-pop track with budots drums that became a nationwide craze in early 2026. SPOT.ph described it playing everywhere from Facebook Reels and jeepney speakers to noontime TV and Zumba tutorials. DZRH reported on 20 March that it had passed 1 million Spotify streams, with Vice Ganda, Charo Santos and the SexBomb dancers among those posting dance versions and a teacher using it as a class energiser. By July, Manny V. Pangilinan had picked it to close an orchestral medley at First Pacific’s 45th anniversary as the song to represent 2026.

Who made it was harder to pin down. SPOT.ph found that the earliest Spotify upload, in 2025, sat under an account called OPM Music Studio whose page did little to show who was behind it, and Spotify did not answer its questions. Online, people pointed to the cover art (gibberish text on a DJ booth, an artificial-looking crowd) as a sign of AI. Some outlets, DZRH among them, went further and described the track as AI-generated without citing any technical evidence. On GMA’s Kapuso Mo, Jessica Soho, producer Gian Vergel said it sounded obviously AI to him.

The man who says he made it disagrees, firmly. Sylvain Hernandez, a French producer based in Leyte who performs as DJ Mogo and is married to a Filipina, told GMA that he made the track with his wife’s help, that it grew out of Leyte cha-cha and the heavy drums of budots, that he built the beat with professional electronic software, and that no part of it was generated by AI. He later told SPOT.ph that he held the earliest verifiable upload, that the earlier date on another account was a metadata entry, and that he had project files and timestamps to show it. He first kept the singer’s identity secret; in April the Daily Tribune reported that he had named a singer, Shane Reyes, on an online show, though without footage of the recording. As of 15 October 2026 nobody has published technical evidence (the tools used, project metadata or a voice analysis) that settles either the AI question or the upload question. This article treats it as what it is: a dispute, not a finding.

In the same news cycle, Pablo of SB19 called the use of artists’ work without consent very unethical and said artists should control how their identity and work are used. FILSCAP, the Filipino songwriters’ and publishers’ society, urged people using generative AI tools to make sure the material they use is licensed.

“Malunggay Pandesal”: a meme with many versions and no clear author. The second case is smaller and murkier. A novelty song about malunggay pandesal, the bread roll with moringa leaves in the dough, circulates on TikTok and YouTube in many versions. One YouTube upload dated 24 February 2026, with about 220,000 views by mid-October, calls itself the original and credits a lyricist. A Spotify release under the same title, dated 10 April 2026, is credited to a different act. I found no news report establishing who wrote the song first, and none of the listings say whether AI tools were used on any version. This article names no author because no reliable source does.

Exhibit 9

	Hawak Mo ang Beat	Malunggay Pandesal
How it spread	Dance craze on Reels and TikTok, in Zumba classes, classrooms and celebrity dance videos	Novelty song about a bread roll, shared on TikTok and YouTube in many versions
Who is credited	Earliest Spotify upload under OPM Music Studio (2025); DJ Mogo (Sylvain Hernandez) says he made it	Different uploads carry different credits; one calls itself the original
What is established	Over 1 million Spotify streams by March 2026; chosen to represent 2026 in an orchestral medley in July	Uploads under the title dated February 2026 (YouTube) and April 2026 (Spotify)
What is disputed or unknown	Whether AI was used; whose upload came first; the singer, named in April without recording footage	Who wrote it first; whether any version used AI
Published technical evidence of AI use	None	None
What the show would say	Name the songwriter and the singer: the story is part of the song	A meme becomes a song when someone can show they made it

Compiled from news reports and platform listings; claims are as reported. This article does not state or imply that either song was made with AI. The last row is the author's reading of the show. The table scrolls sideways on small screens.

Source: SPOT.ph (26 Mar 2026); DZRH (20 Mar 2026); GMA News and GMA Entertainment (27 Mar 2026); Daily Tribune (26 Apr 2026); Philstar (14 Jul 2026); YouTube and Spotify listings

Both cases make the same point. The public could not settle either question by ear, as the 97% blind-test result predicts. What was missing was a credit line: a verified songwriter, a named singer and a disclosure field filled in at upload. On the show’s Mars everyone knew which music came from machines, and the duo won by being visibly themselves. And “Hawak Mo ang Beat” confirms a theme the show keeps returning to: whoever made the audio, the crowd made the hit.

Philippine policy is catching up. In September 2026 the Intellectual Property Office of the Philippines (IPOPHL) put out for consultation draft guidelines on registering AI-related works. Under the draft, using AI does not by itself bar registration, but a prompt alone is not enough human contribution, and purely AI-generated output is not registrable to the extent it lacks human expression. The draft also says registration does not decide whether training on copyrighted material is lawful. In Congress, a May 2026 review by the Foundation for Media Alternatives found all five AI bills it examined still pending, including Senate Bill 758, which would require accounts to disclose AI-generated content and give people exclusive control over their image, voice and identity, and Senate Bill 1714 on digital likeness and deepfakes. For now, labelling AI music in the Philippines depends on voluntary industry labels and platform policies.

What the show saw coming, and what it missed

The second half of the series turns from the music business to politics. Tuesday’s mother, Valerie Simmons, launches a presidential campaign with an anti-immigrant message, calling for an end to the admission of refugees from Earth. A bombing at a weather plant boosts her popularity. As the story closes, musicians who oppose her proposed restrictions on freedom of speech are being arrested, and Carole and Tuesday propose that musicians from across Mars sing together. On New Year’s Eve a crowd of artists, rivals included, gathers at the Mars Immigration Memorial Hall to perform the duo’s new song, “Mother”, in an event the show calls the “Seven-Minute Miracle”.

Along the way the show stages most of 2026’s arguments in miniature.

Exhibit 10

	In the show	In 2026	The open question
Manufactured stars	Producer Tao uses AI to build Angela into a hit singer	Spotify badges AI Persona profiles and keeps them out of recommendations by default	Who is the artist when the face and voice are generated?
Gatekeepers	A star DJ scoffs at music not made by AI	AI is over half of Deezer uploads on peak days but 1–3% of streams	Who decides what gets discovered?
Fraud	A robot video director turns out to be a scam; later, an AI manager makes off with a DJ’s money	Up to 85% of streams on AI tracks at Deezer were fraudulent in 2025	Who pays when automation drains the royalty pool?
Charts and contests	Mars’ Brightest and the Mars Grammys make careers	ARIA bars wholly AI tracks from its charts and awards from August 2026	Should charts measure songs or the people behind them?
Process	A veteran producer makes the duo record one song 52 times	Listeners rate identical audio lower when told an AI performed it	Does how music was made matter if nobody can hear it?
Voice	Algorithms shape every syllable Angela sings	ELVIS Act in force; NO FAKES Act cleared Senate committee in June 2026	Who owns a voice, and for how long?
Stake	Musicians are arrested for speaking out; the finale is a collective performance	Artist groups demand consent before their work enters AI deals	Can a generated song carry the risk that gives a protest song its weight?

Framework by the author. Show details from the official synopsis, Netflix's episode guide, ANN and Wikipedia's episode summaries; 2026 details from the sources cited in this article. The table scrolls sideways on small screens.

The finale makes the stake argument explicitly. The “Seven-Minute Miracle” matters in the story not because the song is well produced, but because the people singing it are visible, named and taking a risk in a political climate where musicians are being arrested. No generator can supply that. A protest song’s force comes from the fact that someone stood behind it.

Read this way, the AI theme and the political theme are the same theme: who gets to be heard, and who is treated as replaceable.

Where the show fell short as a forecast is just as instructive:

1. It imagined AI as a studio, not an app. On Mars, a few powerful producers run the machines. On Earth, more than 100 million people have used a single generator, and the problem is volume, not a handful of synthetic stars.
2. It imagined polish, not spam. The show’s AI hits are good. The real flood is mostly low-effort content and fraud, and the damage is to the royalty pool rather than to anyone’s ears.
3. It skipped training. Nobody on Mars asks what Tao’s AI learned from. On Earth, that is the question in every courtroom from Boston to Munich.
4. On Mars, everyone knew. In the show, AI music is the open default. On Earth, 97% of listeners cannot tell, which is why the defining fight of 2026 is over labels rather than over taste.

Counterpoints

Implications by stakeholder

Exhibit 11

	Artists	Labels and platforms	Listeners	Policymakers	Builders
The lesson from the show	You win on stage and in story, not on polish	A feed of perfect hits still leaves fans wanting someone real	The show gives its big moments to songs the duo wrote themselves	The show’s crisis was political, not technical	Tao’s system made hits; the show asks who answers for them
The risk now	Voice cloned, catalogue licensed without consent	Royalty pools diluted by fraud; trust lost to fake personas	Not knowing what you are listening to	Rules written after the market sets	Products built on unlicensed data that courts unwind
The move	Register your voice rights; disclose your own AI use; invest in live	Adopt AI labels; police fraud; get creator consent, not just label sign-off	Use labels and filters; pay for what you value	Pass digital-replica rights; require disclosure; keep human authorship	Train on licensed data; watermark and credit by default

Framework by the author. The table scrolls sideways on small screens.

For artists

Your sound is no longer your moat on its own; your identity, audience and presence are. Say clearly how you use AI, so that “human-made” is a fact you can point to, and read any letter about AI deals before you opt into anything.

For labels and platforms

The flood is a fraud and trust problem before it is a creative one. The platforms that keep listeners’ trust will be those whose labels are accurate and whose recommendations do not quietly fill with synthetic personas. Licensing with labels is necessary; consent from creators is the next fault line.

For listeners

You already vote with your attention. Use the labels and filters as they arrive, and if you value an artist’s story and stage, spend where it reaches them: tickets, merchandise and direct purchases.

For policymakers

A workable settlement is visible: human authorship for copyright, rights over digital replicas of voice and likeness, standardised disclosure, and enforcement against royalty fraud. Training consent remains open, with courts in different countries reaching different answers. In the Philippines, IPOPHL’s draft guidelines already draw the human-contribution line for copyright registration; the disclosure and likeness bills pending in Congress are where labelling could become law.

For builders

Suno’s own pivot to licensed data, artist opt-ins and download limits shows where this ends up. Build that way from the start: credit inputs, label outputs, and make it easy for an artist to say no.

What to do next: a checklist

If you make music

1. Write down, for each release, what was human-made and what used AI, and submit it through your distributor’s AI credit fields where they exist. Keep your project files and upload records: the “Hawak Mo ang Beat” dispute shows they may be your only proof.
2. Check what your label or publisher has agreed on AI, and ask whether your work is included in any training licence.
3. Protect the layers AI cannot copy: build a mailing list, play live, and make your story visible.

If you run a label, distributor or platform

1. Adopt the IFPI-led AI-Generated and AI-Assisted labels at track level.
2. Measure the share of your royalty pool going to tracks flagged as AI, and the share of those streams flagged as fraud.
3. Ask creators before their work enters an AI deal, and publish how the revenue is split.

If you build AI music tools

1. Use licensed training data, and be able to show it.
2. Watermark outputs and pass disclosure metadata downstream by default.
3. Block prompts that name living artists unless they have opted in.

If you just listen

1. Turn on AI labels and filters where your service offers them.
2. Go to one more gig this year than last year.

Limitations and methodology

• One platform’s data. The upload and stream figures come from Deezer, which is smaller than Spotify or Apple Music, relies on its own detection tool and sells that tool commercially. Other services may see different patterns; most do not publish comparable data.
• Company figures. Suno’s subscriber, revenue and user numbers are the company’s own claims or reports of its investor materials, not audited figures. Live Nation’s figures are its own results.
• Survey limits. The Deezer/Ipsos survey was an online poll of adults aged 18–65 in eight countries, commissioned by a company with an interest in the result. A blind test with three tracks measures one kind of detection, not all of them.
• Small experiments. The listener-bias studies used modest samples (120 people in the performance study) and specific genres. One found no effect for electronic music. The bias is real but conditional.
• Live music is not proof of cause. Record attendance is consistent with value moving to the stage, but live growth predates AI music and is concentrated among large acts.
• Law in motion. Most of the cases cited are unresolved or open to appeal, settlement terms are confidential, and the NO FAKES Act has not passed. Everything here is as of 15 October 2026.
• The show is fiction. Plot details come from the official synopsis, Netflix’s episode guide, ANN’s review and published episode summaries. I have paraphrased scenes rather than quoting dialogue or lyrics, and I have left out plot points I could not verify.
• The Philippine cases are disputes, not findings. Reports on “Hawak Mo ang Beat” conflict on who uploaded it first, whether AI was used and who sang it, and no technical evidence has been published either way. Some outlets described it as AI-generated without citing evidence; I have not repeated that as fact. For “Malunggay Pandesal” I found no news coverage at all, so that account rests only on platform listings, and I name no author. The status of Philippine bills is as of a May 2026 review.
• The framework is mine. The five layers are a thinking tool, not a validated model.

The song still needs someone to sing it

Carole & Tuesday imagined a world where machines could write a perfect hit, and then spent its whole run showing that a perfect hit was not the point. Seven years on, the machines can write the hit. Half the songs arriving at one streaming service on its busiest days are made by them, and almost nobody can hear the difference. What they still cannot do is stand on a stage, carry a story that is true, or put their name to a risk.