Category: AI Transparency

Your Music, AI, and the Machine Unlearning Research Hub: What Every Musician Should Know

If you’re a recording artists or musician, you have most likely heard some form of this pitch:

“Don’t worry. If you don’t like how your music is used for AI, we can always take it back out later.”

That sounds fair.

The hard part is that, today, it is not that simple. And it’s so not simple, that statement may as well be untrue.

That is why we created the  Machine Unlearning Research Hub  at MusicTechPolicy.com—to help musicians understand what machine unlearning is, what it can do today, and what it may be able to do tomorrow.

Think of it like baking a cake

Suppose you bake a cake with flour, eggs, butter, milk, and sugar. The eggs make the cake what it is.

Once the cake comes out of the oven, you cannot remove just the eggs. AI training works in much the same way.

When an AI developer “trains” a model (often on illegally acquired recordings as we are seeing in the litigation), it does not simply store copies of recordings in a folder. It changes billions—or even trillions—of internal values based on every recording it has been trained on and extracted from those training tracks or its “training pipeline”. Each recording and the values the AI extracts from the recording helps shape the finished model.

That’s why it knows what you mean when you ask it to play an Eric Clapton solo or a John Bonham drum fill. That is also why taking one recording back out later is so difficult.

Surely, you say, they didn’t build a dataset that cannot be corrected and only gets larger with human personal rights and copyrights? Well, they kind of did. Not that different from Google Street View. There’s a long “Hotel California” tradition with these people.

So what is machine unlearning?

Machine unlearning is a field of research that asks a simple question:

Can an AI system forget what it learned from a particular piece of data without rebuilding the entire model from scratch?

The quick answer today is not very easily, no matter what they tell you. Researchers all over the world are working on that question.

The results are a mixed bag. A cynic—who me?—might say they’re not trying very hard. But some techniques show real promise. Others work only under very limited conditions with largely toy datasets.

For today’s largest music models, there is still no widely accepted way to guarantee that every artifact of one recording—or one artist—has been completely removed.

That does not mean machine unlearning will never work. It means it is still a research problem.

How much can it fix today?

The honest answer is:

Some things. Not everything.

Researchers have developed methods that can reduce a model’s reliance on particular data or make it much less likely to produce certain results. That’s not 100% or zero depending on how you look at it. But it’s better every year and those advances matter.

But reducing influence is not the same thing as proving that a recording has been fully removed from a commercial AI model. For musicians, that difference is obviously important.

Will it get better?

Almost certainly.

Machine unlearning is receiving significant attention from universities, technology companies, and governments. Why? Well, for one thing there are countries that have the “right to be forgotten” that Google fought so hard against in Europe and especially Germany. There’s also other privacy laws that all of the AI models are essentially out of compliance with, so it’s only a matter of time until there’s a real effort to do something about it.

Five years from now, today’s methods will almost certainly look primitive. Ten years from now, the tools may be far better still. But no one can honestly promise today that future improvements will solve every problem created by yesterday’s training.

That is why decisions made now still matter. Let’s be honest, the best way to stay out of AI is to never be included in AI.

What does this mean if someone wants to license your music?

Ask yourself one simple question:

If I change my mind in three years, how will my music come back out?

Do not settle for vague promises.

Ask specific questions.

  • Will my recordings be kept in a separate training set?
  • Can they be removed without rebuilding the model?
  • What machine unlearning process will be used?
  • Has that process been independently tested?
  • How will you prove my music was removed?
  • What happens if the technology cannot fully remove it?
  • Who pays to remove it?

If the company cannot answer those questions today, it is worth understanding that they are asking you to rely on technology that may not yet exist in the form they hope it will. Or as Blanche Dubois said in Streetcar Named Desire, “I have always relied on the kindness of strangers.” Feel good yet?

That does not necessarily mean you should refuse every AI license.

It does mean you should know exactly what risk you are accepting.

This is why we built the Machine Unlearning Research Hub

Machine unlearning is moving quickly, and new research papers appear almost every week. Some represent genuine breakthroughs, others improve only narrow parts of the problem.

Our goal is not to tell musicians what decision to make. Our goal is to help musicians make informed decisions based on the best available evidence. If someone asks you to license your music for AI, you deserve to know one thing before you sign:

Can they realistically give it back if you ask?

Today, that question deserves a careful answer—not a marketing slogan. That is what the Machine Unlearning Research Hub is here to explore.

by Trichordist

The AI “License” That Isn’t: A Musician’s Checklist for Spotting a Covenant Not to Sue Censorship Trap in Disguise

If you saw the adverb “forever” in a contract, would it give you pause?  It would me—I would wonder who allowed that to slip through.  As text-to-audio generative AI models desperately try to normalize their shoot-ready-aim ingestion of likely stolen works to support their next round of financing (or in some cases perhaps a VC exit by IPO), it’s likely that we will see some efforts at “licensing” reminiscent of the Napster era. “Licenses” that actually paper over the main act—stop the lawsuit before they start. What will that look like and what should we look for buried in the not so fine print?

Generative AI companies are coming to the negotiating table with “music licensing agreements” (and probably other artists’ too). That sounds encouraging — it suggests they want to do the right thing and pay for the content they use. Trust me, they don’t. What they want is to get away with it.

Not every document titled “license” actually functions like one. Some of these deals are better described as litigation insurance dressed up in licensing language, and if you sign one without reading the small print, you may be giving away far more than you realize. Because as Tom Waits taught us in the classic Step Right Up, the large print giveth and the small print taketh away.

Here’s a practical checklist of red flags to watch for if you want to look past the hype:

☐ Check Whether the “License” Actually Includes a Perpetual “Covenant Not to Sue”

A quick definitional note: a covenant not to sue is a contractual promise in which you agree in advance never to bring a legal claim against the other party — regardless of what happens, regardless of whether you even know you have a claim, and often regardless of whether the conduct at issue would otherwise be something you could sue for or get a court to stop (like with an injunction). Unlike a license, which grants permission and can expire, a covenant not to sue can be a permanent waiver of your right to seek a remedy. It’s the difference between handing someone a key to your house and signing a contract promising you’ll never call the police no matter what they do inside.

A real license gives someone permission to use your work under defined conditions, and when the term ends, so does the permission. Watch out for agreements that include a separate clause in which you irrevocably promise — both during and after the term — never to bring any claim against the company or even its users. If the covenant not only covers copyright, publicity rights, moral rights, defamation, and lawsuits over issues that may not exist yet, you are not licensing your work. You are surrendering your right to enforce your rights, permanently. Or you know, “forever.” Make sure you understand the difference before you sign.

☐ Beware a “Non-Exclusive” License That Binds Your Co-Writers

Remember the 100% licensing debacle back in 2016? Tenants-in-common co-ownership of a copyright means any co-owner can grant a non-exclusive license without the others’ consent — but that principle can be weaponized. If an agreement requires you to represent and warrant that no third-party consent is needed, and that you will not encourage your fellow songwriters to take any position inconsistent with the deal (again with the censorship), you may be dragging your co-writers into an arrangement they never agreed to. Worse, if the covenant not to sue extends to all contributors, your co-writers’ enforcement rights could be compromised by a deal they had no say in. Before you sign, make sure the AI platform isn’t leveraging TIC principles to bind people who aren’t at the table. Here’s why this matters in practice: under TIC co-ownership, a non-exclusive license granted by one co-owner is generally binding on all co-owners, even without their consent. That means if you grant a non-exclusive license — or, more critically, a covenant not to sue — covering the full composition, your co-writer may be unable to bring an infringement claim against the same company for the same use, because the company can point to your grant as a complete defense. The co-writer’s right to sue isn’t technically extinguished, but it’s rendered practically worthless. The result is that one publisher’s signature can neutralize enforcement rights across an entire song’s ownership chain. Ask your lawyer.

☐ Look for a Sweeping Pre-Signing Release

Some agreements bury a broad release of all claims arising before the effective date — known and unknown, suspected and unsuspected. That’s not a license, it’s a preemptive settlement masquerading as a license.  Worse, they may ask you to preemptively waive protections like California Civil Code Section 1542, which exists specifically to protect people in this exact situation. The statute prevents people from accidentally releasing claims they don’t yet know about. If an AI company has been training on your catalog without permission for years, a release like this could function like a contractual safe harbor and eliminate any leverage you had to seek compensation for that unauthorized use, or better yet an injunction. You know, kind of like the Music Modernization Act.  Ask yourself: why does a forward-looking license need a backward-looking release? Aside from wanting a pony.

☐ Make Sure “Training Rights” Aren’t Irrevocable in Practice

A license to include your works in a training corpus sounds civilized, but look closely at the removal mechanics (and take a class in “machine unlearning“). If the company is only required to retrain its model a limited number of times per year — say, twice — then even if you pull your catalog, your works may remain embedded in the AI model for months or even “forever”. There are no guarantees that “retraining” will actually remove your works from the model and, in fact, the literature suggests it won’t. And if the covenant not to sue survives the term for anything created while the deal was active (or before the deal was active), the training that already happened is effectively locked in. You may be inadvertently granting a right for yourself and maybe your co-writers that you can never meaningfully take back.

☐ Don’t Pre-Approve a Statutory Streaming License They May Not Be Entitled To

Some AI music platforms aspire to become full on-demand streaming services — and their agreements may include language contemplating exactly that expansion, with provisions stating the company will simply obtain a blanket mechanical license from the Mechanical Licensing Collective under 17 U.S.C. § 115(d). But whether a generative AI platform that creates synthetic music qualifies for the statutory compulsory license designed for traditional on-demand streaming is a genuinely open legal question. If you agree to terms that treat this expansion as a foregone conclusion — or worse, if your covenant not to sue or conditional authorization greenlights streaming functionality by default — you may be conceding the argument before it’s ever litigated. Don’t let an agreement’s assumptions about statutory eligibility become your assumptions. That question should be tested, not waived.

There’s a deeper problem here, too. The Section 115 compulsory license has always contained an anti-piracy prerequisite: you cannot obtain a compulsory license to use a musical composition if the sound recording you’re working from was not lawfully fixed or authorized by the sound recording’s copyright owner. If an AI platform trained its model on sound recordings it ingested without authorization (as is currently being litigated), its entire statutory license theory may be built on a foundation of infringement. The compulsory license was never designed to launder unauthorized use of sound recordings into lawful use of the compositions they embody. By agreeing to terms that treat the platform’s eligibility as settled, you may be implicitly conceding that the platform’s use of those sound recordings was authorized — a concession likely worth far more than anyone is probably paying for it and that can get you sideways with the sound recording owners.

☐ Scrutinize “Guardrail Failure” Safe Harbors

Technical safeguards — input filters, output filters, vocal classifiers — are only as good as the consequences for failure. Be wary of provisions that excuse the AI company from breach liability when its guardrails fail, so long as the failures are characterized as “inadvertent” and “de minimis” which are in the eye of the beholder and just buying a lawsuit.  If the agreement treats guardrail failures as non-breaches by default, you’ve effectively agreed that the company can produce unauthorized outputs of your works without meaningful accountability, as long as it promises to try to fix the problem after the fact with no stick if it fails.

☐ Watch Who Owns the AI Model — and Learned “Insights”

Ownership clauses in AI agreements often go further than you’d expect. The company may claim ownership not just of the AI model, but of all “insights” it learns which can be damn near anything. That “learning” is including those developed using your content. That language could be read to mean that everything the AI extracts, learns, or derives from your catalog belongs to the platform. Make sure you are not inadvertently ceding ownership of the creative intelligence embedded in your works.

☐ Don’t Let a Third Party’s Binding Calculation Determine Your Payment

I am no fan of market share revenue share deals, particularly when the revenue pool is a fixed number. Some revenue pool deals delegate the calculation of your market share — and therefore your payment — to a third party, and then declare that calculation to be binding on you, even if you had nothing to do with calculating either your market share (under the “license” not in general) or your share of the revenue or minimum guarantee. If you have no right to challenge the methodology or the math, or to audit that third party, you could be found to have agreed to be paid whatever someone else decides you’re owed, with no recourse if they get it wrong.

☐ Trace the Revenue Through Every Deduction Before Celebrating the Rate

A headline royalty rate means nothing if it’s applied to a revenue base that has already been carved down by layers of deductions including off the top fees, advertising costs, technical fees, inference compute costs, app-store commissions, performance royalties, and more. Each deduction may sound reasonable in isolation, but stacked together, they can reduce the revenue pool to a fraction of what you’d expect. Always model your actual payout, not just the stated percentage.

☐ Beware Conditional Authorizations That Expand the Deal Without Your Active Consent

Some agreements include provisions for future features such as API access, enterprise customer pricing, new product versions, downloads, or interactive streaming that automatically activate once a majority of other rightsholders sign on. If the trigger is what other labels or publishers do rather than what you approve, you may find that the scope of the license has expanded well beyond what you agreed to, without any additional negotiation or compensation.  And remember what they say about if everyone else was running off a cliff.

☐ Censorship Clauses: Don’t Agree to Never Contradict the Deal

Representation and warranty sections sometimes include a covenant that you will not take, or encourage your artists or songwriters to take, any position at any time (whether during or after the term and whether or not truthful) that is inconsistent with the hoorah narrative about AI or the applicable license. Read that carefully. It could be interpreted to prevent you — or your writers — from ever publicly criticizing the deal, advocating for stronger protections, or supporting legislation that conflicts with the agreement’s framework, even years after it expires even if the AI platform is in breach.  Aside from censoring your freedom of speech, this has nothing, and I mean nothing, to do with a license.

☐ Understand What Walled Garden Actually Means for Your Writers

A service that restricts AI-generated content to a “walled garden” with no downloads sounds contained. But if users can generate unlimited content using your works, share tracks or links to tracks across social media, and the company retains the right to expand into other product lines (including non-AI product lines like an on-demand streaming service) all within the same agreement, any “walled garden” may be pretty ephemeral.

☐ Confirm That Your Minimum Guarantee Is Actually Guaranteed

A minimum guarantee that is “fully recoupable” against future royalties is not a floor — it’s an advance. If the service underperforms, you keep the advance, but if it succeeds, the guarantee is just an interest-free loan. Make sure you understand whether the guarantee represents real minimum compensation or simply front-loaded royalties you would have earned anyway. Also, plan for both failure and success—if you’re a publisher or label, how on earth are you going to be able to account to your songwriters or artists while you’re recouping any minimum guarantee or afterwards?

☐ Demand Flow-Down Protections in the Platform’s Terms of Service

An AI platform’s Terms of Service are where your contractual protections actually meet the end user — and if the ToS doesn’t carry your rights forward, your rights may exist only on paper, if at all. Before you sign, confirm that the agreement requires the platform’s user-facing ToS to include, at a minimum:

  • An ownership disclaimer — users must acknowledge they acquire no copyright or ownership interest in AI outputs that embody your works.
  • A reverse-engineering prohibition — users must be prohibited from extracting, reconstructing, or isolating your works from the model’s outputs.
  • A downstream training restriction — users must be barred from using AI-generated outputs containing your works to train their own AI models.
  • User indemnification that flows to you — if a user misuses your works, you shouldn’t have to rely solely on the platform to make you whole.
  • A meaningful commercial-use definition and enforcement mechanism — if the deal says “personal, non-commercial use only,” the ToS needs to define what that means and impose real consequences for violations, not just account revocation. Remember, “non-commercial” has been used for text and data mining exceptions in various countries that are huge and unintended exceptions to copyright.
  • An anti-circumvention clause — users should be explicitly prohibited from attempting to bypass input/output filters and guardrails, not just prevented by technology that may fail.
  • Publisher review and approval rights over ToS language — if the agreement references ToS protections as part of your deal, you should have the right to review and approve the actual language implementing them.

If the platform isn’t required to flow these protections down to users, then the guardrails in your agreement are a ceiling, not a floor.

☐ Ask Whether Any Other Rightsholders Are Getting Equity — and Whether You Are, Too

If the agreement includes an MFN clause promising you’ll receive the “most favorable economic terms” offered to any other licensor, ask the obvious follow-up: are large rightsholders receiving equity stakes in the AI platform as part of their deals like the majors and Merlin did with Spotify? We’ve seen this movie before. When streaming platforms launched, labels negotiated equity positions that dramatically increased the total value of their agreements — value that was never shared with publishers or songwriters. If an AI company is offering stock, warrants, or other equity consideration to labels while offering publishers only cash royalties and a minimum guarantee, then the MFN clause is cosmetic. The “most favorable economic terms” aren’t favorable at all if they exclude the most valuable component of the deal. Before you accept the premise that you’re being treated equally, ask what the royalty rate would look like if nobody were getting equity. That’s the number that tells you whether the cash terms are fair on their own — or whether they’re subsidized by equity you’ll never see.

Now What: When an AI company presents you with a “licensing agreement,” read it like a litigator, not a dealmaker. (When presented with a contract, dealmakers look at the money, litigators look at the remedies.). The title of the document matters far less than what’s inside it. If the agreement includes a perpetual covenant not to sue, a retroactive release of claims, and an ownership clause that captures everything the AI learns from your work, what you’re being offered isn’t a license — it’s capitulation with a royalty attached.

Know what you’re signing. Read it yourself, don’t buy the hoorah.

by thetrichordist

Yes, it really is a data center next to the Nashville Zoo

The proposal to build a massive data center adjacent to the Nashville Zoo raises a simple question: Have we completely lost our sense of priorities?

Nashville’s zoo exists to provide education, conservation, recreation, and a rare connection between people and animals. Families bring children to experience living animals, open space, and a respite from the relentless industrialization that increasingly consumes American communities. Yet now residents are being told that one of the city’s most treasured public assets should coexist with an industrial-scale computing facility whose primary purpose is to feed the endless demand for artificial intelligence, cloud computing, and speculative digital services. This is insane and it is exactly backward.

The burden of proof should not fall on citizens to explain why they do not want a hyperscale data center next to a zoo. The burden should fall on developers to explain why a project requiring enormous quantities of electricity, water, backup generation, transmission infrastructure, truck traffic, and round-the-clock industrial operations and both light and noise pollution belongs there in the first place.

The economic promises attached to these projects are increasingly difficult to take seriously as has been demonstrated by a recent study of data center job impact in Texas. Across the country, data center developers routinely advertise billions of dollars in investment while generating surprisingly few permanent jobs. Independent research has repeatedly found that many large data centers produce limited long-term employment relative to their physical footprint, utility demands, and public subsidies. Communities are often left with the costs while investors and distant technology companies capture the benefits.

Meanwhile, the impacts are immediate and local.

Residents face years of construction activity, noise, traffic, and visual blight. Wildlife habitats are disrupted. Open space disappears. Transmission lines, substations, backup generators, and supporting infrastructure permanently alter the character of surrounding neighborhoods. Once built, these facilities are effectively impossible to remove. They become permanent industrial fixtures.

The Nashville Zoo should not become collateral damage in the AI arms race.

Even more troubling is the uncertainty surrounding the long-term economics of artificial intelligence itself. Technology companies are spending hundreds of billions of dollars based on forecasts that extend years into the future. Yet many of the underlying assumptions remain unproven. No one can say with confidence what demand for AI services will look like five, ten, or twenty years from now. If those forecasts prove wrong, communities could be left staring at the digital equivalent of abandoned factories—massive, energy-hungry facilities built for demand that never materialized.

The risk is not theoretical. Economists have a name for this phenomenon: stranded assets.

A zoo is a long-term civic investment. It creates educational, environmental, and cultural value that can endure for generations. A speculative AI data center is a bet on future demand forecasts generated in corporate boardrooms and venture-capital presentations.

When those two visions collide, the choice should not be difficult.

Nashville should protect its zoo, its surrounding communities, and its quality of life. There are countless locations better suited for industrial-scale computing infrastructure. A zoo is not one of them.

Some places should remain places for people, families, wildlife, and conservation. Not every acre of land needs to be sacrificed to the next technological gold rush.

The Nashville Zoo deserves better than becoming the neighbor of a machine. And believe me, if they’ll do it in Nashville they’ll do it anywhere. The Zoo has a Change.org petition you can sign if you agree.

by Trichordist Editor

The Growing Backlash Against AI Data Centers: Local Resistance and the Infrastructure Crunch

As we’ve reported many times, communities across the US are increasingly pushing back against the explosive growth of AI-driven data centers. Major concerns include skyrocketing electricity demand, massive water consumption for cooling, noise pollution from giant fans, loss of prime agricultural and residential land, and rising utility bills passed on to local residents. As of May 2026, independent trackers report approximately 69–78 U.S. jurisdictions that have enacted bans, restrictions, or moratoriums on new data centers. Many of these measures also target the new high-voltage transmission lines required to power them.

This wave of resistance highlights a deepening tension between the rapid expansion of AI infrastructure and local priorities around quality of life, sustainability, and community control.

1. Michigan: The Epicenter of Local Moratoriums

I think you could safely say that Michigan currently leads the nation in local opposition to data center construction, largely triggered by the controversial $16+ billion OpenAI-Oracle Stargate AI data center project in Saline Township, Washtenaw County. Despite a 4-1 township planning commission vote against rezoning and strong resident protests, the Stargate construction project advanced through legal channels, igniting widespread defensive actions across the state.

  • More than 50 communities (cities and townships) have enacted temporary moratoriums, covering roughly 1,500 square miles — an area comparable to the size of Rhode Island.
  • Between 25 and 51 active local moratoriums are in place as of early 2026.
  • State lawmakers have introduced bills (HB 5594–5596) calling for a one-year statewide pause on new hyperscale data centers, along with stricter rules on water and electricity connections.
  • Some utilities, such as Ypsilanti, have imposed their own 12-month bans on water hookups for large AI facilities—but that will eventually expire.

Key issues in Michigan should sound familiar: massive water usage, strain on the electrical grid, and the loss of local zoning authority.

2. Virginia: Transmission Line Battles in “Data Center Alley”

Virginia is home to the highest concentration of data centers in the United States (over 550 facilities), particularly in Northern Virginia. Opposition here focuses heavily on both the data centers and the extensive transmission lines needed to support them.

  • Strong protests in Loudoun, Prince William, Hanover, and other counties against new projects and expansions.
  • Major conflicts over high-voltage lines such as the Valley Link and Joshua Falls projects, which cross multiple counties and impact neighborhoods, historic sites, and conserved rural land.
  • Dominion Energy has faced repeated legal and community challenges regarding route selections.
  • Legislative debates continue over ending billions in tax incentives and studies projecting residential electricity rate increases of up to $37 per month by 2040.
Breakfast at Buck’s of Woodside—if you’re not at the table you are on the menu

3. Georgia: Statewide Pause Efforts Amid High Project Volume

Georgia has seen hundreds of announced data center projects, prompting both local and statewide responses.

  • Bills such as HB 1059 and HB 1012 propose temporary statewide pauses on new permitting (potentially until 2027–2028) to allow time for impact studies.
  • Several counties, including DeKalb and Camden, have passed moratoriums ranging from several months to a year while updating zoning ordinances.
  • Residents voice concerns about energy costs, water consumption, loss of land, and whether tax incentives truly benefit local communities.

Georgia’s combination of legislative proposals and county-level actions reflects growing resistance in a rapidly developing market.

4. North Carolina: Rising Local and Policy Pushback

North Carolina ranks among the top states for new moratorium activity as data center developers expand beyond traditional East Coast hubs.

  • Multiple counties and municipalities have passed restrictions or temporary moratoriums citing infrastructure strain, zoning issues, and community impacts.
  • Policy proposals such as HB 1063 seek to require hyperscale developers to fully cover the costs of power, water, and grid upgrades rather than passing them to ratepayers.
  • Growing focus on the environmental and visual effects of both data centers and supporting transmission lines.

North Carolina represents an emerging hotspot where early local actions may shape future statewide policy.

5. Indiana: County-Level Resistance and High-Stakes Conflicts

Indiana has seen intense localized opposition, particularly in rural counties.

  • Counties such as White and Fulton have enacted 6-to-12-month moratoriums to study impacts and strengthen local ordinances.
  • Trackers show at least 6 formal actions, with several others in discussion.
  • Primary concerns include the conversion of prime agricultural land, rising utility rates, and the industrialization of rural communities.

Indiana illustrates how even mid-sized proposals can trigger strong community responses and political tension.

Broader Implications and the Path Forward

The five most active states — Michigan, Virginia, Georgia, North Carolina, and Indiana — capture the national picture. Resistance is bipartisan, spans urban and rural areas, and increasingly includes opposition to the massive transmission lines that accompany data center projects.

Common themes include fears that data centers consume disproportionate amounts of power and water while shifting costs onto existing residents. Proponents argue these facilities bring jobs, tax revenue, and are essential for America’s AI competitiveness. Critics insist that growth must be responsible, with full cost recovery, better siting practices, efficiency standards, and genuine community input.

As AI demand continues to surge, this local “revolt” tests whether the physical infrastructure can scale fast enough without compromising quality of life and environmental goals. I think the national consensus is a big no.

Expect more moratoriums, ballot initiatives, legal battles, and negotiations in the coming months. The outcome will significantly influence not only the future of AI but also national energy policy and land-use planning for years to come.

by thetrichordist

@RonanFarrow and @AndrewMarantz: Sam Altman May Control Our Future—Can He Be Trusted?

Ronan Farrow and Andrew Marantz investigate Sam Altman’s leadership of OpenAI, based on internal documents and more than 100 interviews. They center on a core tension: Altman has positioned himself as a steward of humanity’s most powerful technology, yet many colleagues and insiders question whether he can be trusted with that responsibility. Internal memos compiled by senior figures—including chief scientist Ilya Sutskever—allege a pattern of misleading statements and evasiveness, particularly around AI safety and governance.  Shocking, ain’t it?

The piece traces OpenAI’s evolution from a nonprofit founded to prioritize safety over profit into a commercially driven company pursuing massive scale and valuation. Along the way, Altman is portrayed as highly ambitious, politically savvy, and willing to push boundaries—sometimes at the expense of transparency or institutional safeguards. 

It also situates these concerns within the broader stakes of artificial general intelligence: if such systems emerge, the individuals controlling them could wield unprecedented global power. The article ultimately raises an unresolved question—whether the rapid centralization of technological authority in a single leader and company is compatible with the level of trust and accountability that such power demands.

Read it on the New Yorker.

by thetrichordist

Say No to Suno

Late last year, thieves disguised as construction workers broke into the Louvre during broad daylight, grabbed more than $100 million worth of crown jewels, and roared off on their motorbikes into the busy streets of Paris. While some of those thieves were later arrested, the jewelry they stole has yet to be recovered, and many fear those historic works of artistry have already been recut, reset, and resold.

Closer to home, but no less nefarious, is the brazen rip-off of artists enabled by irresponsible AI, whose profiteers are recutting, remixing, and reselling original works of artistry as something new.  The hijacking of the world’s entire treasure-trove of music floods platforms with AI slop and dilutes the royalty pools of legitimate artists from whose music this slop is derived. 

Meanwhile, those who are promoting this new business model are operating in broad daylight, too – minus the yellow safety vests.  That is AI music company Suno, the brazen “smash and grab” platform whose “Make it Music” ad campaign suggests that the most personal and meaningful forms of music can now be fabricated by their unauthorized AI platform machinery trained on human artists’ work. 

How significant is this activity?  Publicly revealed data says Suno is used to generate 7 million tracks a day, a massive quantity that suggests a dominant market share of AI tracks.  According to recent reports, Deezer “deems 85% of streams of fully AI-generated tracks [on its service] to be fraudulent,” and that such tracks include outputs from major generative models.  As JP Morgan’s analysts said, Deezer’s data “should be indicative of the broader market.”  Suno has yet to demonstrate persuasively that its platform does not, in practice, serve as a scalable input into streaming-fraud schemes — raising a serious concern that Suno has, in effect, become a fraud-fodder factory on an industrial scale.

In a February 2 LinkedIn post, Paul Sinclair, Suno’s Chief Music Officer, claims that his company’s platform is about “empowerment” that enables “billions of fans to create and play with music.”  He argues that closed systems are “walled gardens” that deny people access to the full joy of music.

Ironically, Sinclair’s choice of analogy undermines his own argument.  Ask yourself: just why are most gardens surrounded by fences or walls?  To keep out rabbits, deer, raccoons and wild pigs seeking a free lunch.  We cultivate, nurture and protect our gardens precisely because that makes them much more productive over the long run.

While Sinclair may be loath to admit it, AI is fundamentally different from past disruptive innovations in the music industry.  The phonograph, cassettes, CDs, MP3s, downloads, streaming – all these technologies were about the reproduction and distribution of creative work.  By contrast, irresponsible AI like Suno appropriates and plunders such creative work while undermining the commercial ecosystem for artists.

Think back to the days of Napster.  What brought the music industry back from the ruinous abyss of unfettered digital piracy?  It was the very “closed systems” that Sinclair derides as exclusionary.  At least streaming platforms maintain access controls and content management systems that enable creator compensation, even if the economic outcomes for many creators remain inadequate.  Should we be against Apple Music, Spotify, Deezer, YouTube Music, and Amazon Music?  What about Netflix, Disney+ and HBO, too, while we’re at it?

At its core, Sinclair’s argument is just a tired remix of the old trope that “information wants to be free.”  What that really means is: “We want your music for free.”

Artists need to understand Suno’s game.  They are not putting technology in the service of artists; they are putting artists in the service of their technology.  Every time artists’ creations are used by the platform, those creations have just unwittingly been contributed to the creation of endless derivatives of artists’ own work, not to mention AI slop, with limited or no remuneration back to the human creators.  Suno built its business on our backs, scraping the world’s cultural output without permission, then competing against the very works exploited.

It’s also important to keep in mind that using Suno to generate audio output calls into question the copyrightability of whatever Suno creates.  Most countries around the world including the US Copyright Office have been clear that generative AI outputs are largely ineligible for a copyright – meaning the economic value of the Suno creation lies solely with Suno, not with the artist using it.  The only ones gaining empowerment from Suno are Suno themselves.

Many in our community are embracing responsible AI as a tool for creation, and as a means for fans to explore and interact with our artistry.  That’s wonderful.  But it’s not the same as creating an environment where AI-generated works sourced from our music are mass distributed to dilute our royalties or, worse yet, reward those actively seeking to commit fraud.  Artists need to know the difference – all AI platforms are not the same, and Suno, which is being sued for copyright infringement, is not a platform artists should trust.

Responsible AI-generated music must evolve within a framework that respects and remunerates artists, enhances human creativity rather than supplants it, and empowers fans to engage with the music they love.  At the same time, AI services must preclude mass distribution of slop and prevent fraudsters from destroying the very ecosystem that has been built to reward and sustain artists and audiences alike.

All of us, including billions of music fans, share an urgent, deep and abiding interest in protecting and rewarding human genius, even as AI continues to change our industry and the world in unimaginable ways.  So in 2026, even as the Louvre continues to revamp its own approach to security, we in the arts must rise to confront those who would “smash-and-grab” our creativity for their own benefit.

Together, while embracing innovation, we must work to establish more effective safeguards – both legal and technological – that better promote and protect all creative artists, our intellectual property, and the spark of human genius.

Say no to Suno. Say yes to the beauty and bounty of the gardens that feed us all.

Signed: 

Ron Gubitz, Executive Director, Music Artist Coalition

Helienne Lindvall, Songwriter and President, European Composer and Songwriter Alliance

David C. Lowery, Artist and Editor The Trichordist

Tift Merritt artist, Practitioner in Residence, Duke University and Artist Rights Alliance Board Member

Blake Morgan, artist, producer, and President of ECR Music Group.

Abby North, President, North Music Group

Chris Castle, Artist Rights Institute

by Trichordist Editor

Synthetic Emotion from The Music Department: Suno’s Unsettling Ad Campaign and the Return of Orwell’s Machine-Made Culture from 1984

In George Orwell’s 1984, the “versificator” was a machine designed to produce poetry, songs, and sentimental verse synthetically, without human thought or feeling. Its purpose was not artistic expression but industrial-scale cultural production—filling the air with endless, disposable content to occupy attention and shape perception. Nearly a century later, the comparison to modern generative music systems such as Suno is difficult to ignore. While the technologies differ dramatically, the underlying question is strikingly similar: what happens when music is produced by machines at scale rather than by human experience?

Orwell’s versificator was built for scale, not meaning (reminding you of anyone?). It generated formulaic songs for the masses, optimized for emotional familiarity rather than originality. Suno, by contrast, uses sophisticated machine learning trained on vast corpora of human-created music to generate complete recordings on demand that would be the envy of Big Brother’s Music Department. Suno can reportedly generate millions of tracks per day, a level of output impossible in any human-centered musical economy. When music becomes infinitely reproducible, the limiting factor shifts from creation to distribution and attention—precisely the dynamic Orwell imagined.

Nothing captures the versificator analogy more vividly than Suno’s own dystopian-style “first kiss” advertisingcampaign. In one widely circulated spot, the product is promoted through a stylized, synthetic emotional narrative that emphasizes instant, machine-generated musical cliche creation untethered from human musicians, vocalists, or composers. The message is not about artistic struggle, collaboration, or lived expression—it is about mediocre frictionless production. The ad unintentionally echoes Orwell’s warning: when culture can be manufactured instantly, expression becomes simulation. And on top of it, those ads are just downright creepy.

The versificator also blurred authorship. In 1984, no individual poet existed behind the machine’s output; creativity was subsumed into a system. Suno raises a comparable question. If a system trained on thousands or millions of human performances produces a new track, where does authorship reside? With the user who typed a prompt? With the engineers who built the model? With the countless musicians whose expressive choices shaped the training data? Or nowhere at all? This diffusion of authorship challenges long-standing cultural and legal assumptions about what it means to “create” music.

Another parallel lies in standardization. The versificator produced content that was emotionally predictable—pleasant, familiar, subservient and safe. Generative music systems often display a similar gravitational pull toward stylistic averages embedded in their training data that has been averaged into pablum. The result can be competent, even polished output that nevertheless lacks the unpredictability, risk, and individual voice associated with human artistry. Orwell’s concern was not that machine-generated culture would be bad, but that it would be flattened—replacing lived expression with algorithmic imitation. Substitutional, not substantial.

There is also a structural similarity in scale and economics. The versificator’s value to The Party lay in its ability to replace human labor in cultural production and to force the creation of projects that humans would find too creepy. Suno and similar systems raise analogous questions for modern musicians, particularly session players and composers whose work historically formed the backbone of recorded music. When a single system can generate instrumental tracks, arrangements, and stylistic variations instantly, the economic pressure on human contributors becomes obvious. Orwell imagined machines replacing poets; today the substitution pressure may fall first on instrumental performance, arrangement, sound designer, and production roles.

Yet the comparison has limits, and those limits matter. The versificator was a tool of centralized control in a dystopian state, designed to narrow human thought. Suno operates in a pluralistic technological environment where many artists themselves experiment with AI as a creative instrument. Unlike Orwell’s machine, generative music systems can be used collaboratively, interactively, and sometimes in ways that expand rather than suppress creative exploration. The technology is not inherently dystopian; its impact depends on how institutions, markets, and creators choose to shape it.

A deeper difference lies in intention. Orwell’s versificator was never meant to create art; it was meant to simulate it. Modern generative music systems are often framed as tools that can assist, augment, or inspire human creativity. Some artists use AI to prototype ideas, explore unfamiliar styles, or generate textures that would be difficult to produce otherwise. In these contexts, the machine functions less like a replacement and more like a new instrument—one whose cultural role is still evolving.

Still, Orwell’s versificator is highly relevant to understanding Suno’s corporate direction. When cultural production becomes industrialized, quantity can overwhelm meaning. The risk is not merely that machine-generated music exists, but that its scale reshapes attention, value, and recognition. If millions of synthetic tracks flood listening environments as is happening with some large DSPs, the signal of individual human expression may become harder to perceive—even if human creativity continues to exist beneath the surface.

The comparison between Suno and the versificator symbolizes the moment when technology challenges the boundaries of authorship, creativity, and cultural labor. Orwell warned of a world where machines produced endless culture without human voice. Today’s question is subtler: can society integrate generative systems in ways that preserve the distinctiveness of human expression rather than dissolving it into algorithmic slop?

The answer will not come from technology alone. It will depend on choices—legal, cultural, and economic—about how machine-generated music is labeled, valued, and integrated into the broader creative ecosystem. Orwell imagined a future where the machine replaced the poet. The task now is to ensure that, even in an age of generative AI, the humans remains audible.

by Trichordist

Stealing Isn’t Innovation!

Don’t let the so-called “AI czar” sell you the idea that changing the law to legalize taking artists’ work without consent is innovation. It isn’t.

Innovation creates new value. The AI boondoggle takes existing value from creators and communities and hands it to a small number of tech companies—without permission, without payment, and without accountability but with a nuclear reactor next to your house.

Artists aren’t raw material. They’re rights-holders under U.S. law. Rewriting those rights to subsidize AI business models isn’t progress—it’s a policy choice to reward theft at scale.

AI can thrive without gutting creative rights. But that requires consent, licensing, and fair compensation—not retroactive immunity dressed up as innovation.

Stealing isn’t innovation. It’s just stealing, with a press strategy.

Find out more at Stealing Isn’t Innovation and @human_artistry

by Trichordist

NYT: Silicon Valley’s Man in the White House Is Benefiting Himself and His Friends

This image has an empty alt attribute; its file name is image-9.png

The New York Times published a sprawling investigation into David Sacks’s role as Trump’s A.I. and crypto czar. We’ve talked about David Sacks a few times on these pages. The Times’ piece is remarkable in scope and reporting: a venture capitalist inside the White House, steering chip policy, promoting deregulation, raising money for Trump, hosting administration events through his own podcast brand, and retaining hundreds of A.I. and crypto investments that stand to benefit from his policy work.

But for all its detail, the Times buried the lede.

The bigger story isn’t just ethics violations. or outright financial corruption. It’s that Sacks is simultaneously shaping and shielding the largest regulatory power grab in history: the A.I. moratorium and its preemption structure.

Of all the corrupt anecdotes in the New York Times must read article regarding Viceroy and leading Presidential pardon candidate David Sacks, they left out the whole AI moratorium scam, focusing instead on the more garden variety of self-dealing and outright conflicts of interest that are legion. My bet is that Mr. Sacks reeks so badly that it is hard to know what to leave out. Here’s a couple of examples:

This image has an empty alt attribute; its file name is image-10.png

There is a deeper danger that the Times story never addresses: the long-term damage that will outlive David Sacks himself. Even if Sacks eventually faces investigations or prosecution for unrelated financial or securities matters — if he does — the real threat isn’t what happens to him. It’s what happens to the legal architecture he is building right now.

This image has an empty alt attribute; its file name is sacks-american-flag.jpg

If he succeeds in blocking state-law prosecutions and freezing A.I. liability for a decade, the harms won’t stop when he leaves office. They will metastasize.

Without state enforcement, A.I. companies will face no meaningful accountability for:

  • child suicide induced by unregulated synthetic content
  • mass copyright theft embedded into permanent model weights
  • biometric and voiceprint extraction without consent
  • data-center sprawl that overwhelms local water, energy, and zoning systems
  • surveillance architectures exported globally
  • algorithmic harms that cannot be litigated under preempted state laws

These harms don’t sunset when an administration ends. They calcify. It must also be said that Sacks could face state securities-law liability — including fraud, undisclosed self-dealing, and market-manipulative conflicts tied to his A.I. portfolio — because state blue-sky statutes impose duties possibly stricter than federal law. The A.I. moratorium’s preemption would vaporize these claims, shielding exactly the conduct state regulators are best positioned to police. No wonder he’s so committed to sneaking it into federal law.

The moratorium Sacks is pushing would prevent states from acting at the very moment when they are the only entities with the political will and proximity to regulate A.I. on the ground. If he succeeds, the damage will last long after Sacks has left his government role — long after his podcast fades, long after his investment portfolio exits, long after any legal consequences he might face.

The public will be living inside the system he designed.

There is one final point the public needs to understand. DavidSacksis not an anomaly. Sacks is to Trump what Eric Schmidt was to Biden: the industry’s designated emissary, embedded inside the White House to shape federal technology policy from the inside out. Swap the party labels and the personnel change, but the structural function remains the same. Remember, Schmidt bragged about writing the Biden AI executive order.

This image has an empty alt attribute; its file name is of-all-the-ceos-google-interviewed-eric-schmidt-was-the-only-one-that-had-been-to-burning-man-which-was-a-major-plus.jpg

So don’t think that if Sacks is pushed out, investigated, discredited, or even prosecuted one day — if he is — that the problem disappears. You don’t eliminate regulatory capture by removing the latest avatar of it. The next administration will simply install a different billionaire with a different portfolio and the same incentives: protect industry, weaken oversight, preempt the states, and expand the commercial reach of the companies they came in with.

The danger is not David Sacks the individual. The danger is the revolving door that lets tech titans write national A.I. policy while holding the assets that benefit from it. As much as Trump complains of the “deep state,” he’s doing his best to create the deepest of deep states.

Until that underlying structure changes, it won’t matter whether it’s Sacks, Schmidt, Thiel, Musk, Palihapitiya, or the next “technocratic savior.”

The system will keep producing them — and the public will keep paying the price. For as Sophocles taught us, it is not in our power to escape the curse.

by thetrichordist