Month: March 2026

AI, Soft Power, and the New Thucydides Trap

The White House’s latest AI framework reads like a familiar story dressed in new clothes: we must move fast, avoid “overregulation,” and ensure that the United States “wins” the AI race—because China.

That framing is not new. It is, in fact, a modern version of the Thucydides Trap: the idea that when a rising power threatens to displace an established one, conflict—economic, political, or otherwise—becomes more likely. But what is striking here is not the invocation of competition. It’s how narrowly that competition is defined.

The framework implicitly treats AI dominance as a function of compute, capital, and model scale. Build bigger models faster, feed them more data, and ensure that domestic firms face as few constraints as possible. In that telling, creators, rights, and consent become secondary considerations—at best friction, at worst obstacles.

But that is a profound misread of where U.S. advantage actually lies.

American leadership has never been just about scale. It has been about legitimacy—the ability to build systems that other countries, companies, and individuals trust enough to adopt. That is the essence of soft power. And soft power is not generated by extraction; it is generated by rules that are perceived as fair.

When U.S. policy signals that training on creative works without meaningful consent is acceptable—or even necessary to “win”—it risks trading long-term legitimacy for short-term acceleration. That is a dangerous bargain. It tells the world that American AI leadership is built not on innovation alone, but on the uncompensated appropriation of global cultural and informational resources.

Other jurisdictions are already responding. The EU is experimenting with transparency mandates. Rights holders globally are pushing for enforceable consent regimes. Even countries that want to encourage AI development are increasingly wary of frameworks that look like data extraction at scale without accountability.

This is where the Thucydides analogy breaks down—or at least becomes more complicated. The real risk is not simply that China catches up technologically. It is that the United States, in trying to outrun that possibility, undermines the normative foundations of its own leadership.

Soft power erosion is not dramatic. It doesn’t announce itself with a headline. It accumulates quietly: in trade negotiations, in regulatory divergence, in the willingness of other countries to align—or not align—with U.S. standards. Over time, that erosion can matter more than any benchmark score or model release.

There is another path. The United States could lead by insisting that AI development is compatible with consent, compensation, and provenance. It could treat creators not as inputs to be harvested, but as stakeholders in a system that depends on their work. It could build infrastructure—technical and legal—that makes those principles operational, not aspirational.

That approach may look slower in the short term. It may impose costs that competitors are willing to ignore. But it is also how durable leadership is built.

Because in the long run, the question is not just who builds the most powerful models. It is who builds systems that the rest of the world is willing to trust.

And that is a competition the United States cannot afford to lose.

The Constitutional Shadow of the White House AI Framework: Law Without Law

One of the most important things about the White House AI framework released last week is what it is not.

It is not an executive order.

That may sound like a technical distinction, but it is doing an enormous amount of work here. Because by avoiding the form of an executive order, the framework avoids something even more important: Judicial review.

An executive order that attempted to declare AI training on copyrighted works lawful—or to constrain Congress from acting—would immediately invite challenge in the very judicial branch the framework also seeks to influence. Oh, that would be fun.

It would raise Administrative Procedure Act questions. It would trigger separation-of-powers scrutiny. It would likely be litigated within days.

This framework does none of that and is not susceptible to judicial challenge.

Instead, it achieves much of the same practical effect—shaping legal outcomes, constraining policy space, and signaling preferred doctrine—without creating a justiciable action. It is, in effect, law without law, and outcomes by positioning. Silicon Valley’s favorite.

Takings by Policy, Not Statute

Start with the most obvious constitutional issue: the Takings Clause of Fifth Amendment of the U.S. Constitution which states that “private property [cannot] be taken for public use, without just compensation.”

Copyright is a form of property. That is not controversial. It is a statutory property right grounded in the Constitution’s Intellectual Property Clause, and it carries exclusive rights that have long been understood as economically valuable.

Now consider what the White House framework does.

It declares that AI training—mass, indiscriminate ingestion of copyrighted works—as lawful. It does so without requiring compensation. And it does so in a context where the resulting systems can substitute for, or diminish the market for, the original works.

If that official policy position of the Executive Branch were enacted into law, it would raise a straightforward question:

Has the government authorized the use of private property for public and commercial purposes without compensation? Or more directly, has the Executive Branch just announced that will not prosecute that indiscriminate ingestion for any reason? Can we expect to see amicus briefs from the Solicitor General opposing copyright owners pursuing their rights in court?

That is sounding a lot like a taking.

But because the framework is not law, it avoids the moment where that question must be answered. It does not extinguish rights formally. It renders them economically hollow in practice, while leaving the formal structure intact.

That is the key move: functional elimination without formal abolition.

Ex Post Facto in Everything but Name

The framework also raises a second, less discussed issue: the logic of ex post facto lawmaking.

The Ex Post Facto Clause technically applies to criminal law. But the underlying principle is broader: the government should not change the legal consequences of past conduct to benefit favored actors or disadvantage others. Of course, copyright owners raising this argument will have the Spotify retroactive safe harbor in Title I of the Music Modernization Act thrown in their face as rank hypocrisy, which they would richly deserve, although as any 10 year old can tell you, two wrongs don’t make a right, at least in theory.

Here, the timeline matters.

  • Massive datasets have already been scraped.
  • Models have already been trained.
  • The conduct that enabled this may, in many instances, have been legally questionable—and in cases of willful infringement, potentially criminal under federal copyright law. Or if you listen to me, the largest case of criminal copyright infringement in history.

Now comes the policy years after the fact in the face of over 150 AI lawsuits all based on copyright infringement to one degree or another:

Training is lawful.

That looks less like interpretation and more like retroactive validation.

Even if framed as civil doctrine, the effect is similar to retroactive decriminalization of conduct tied to vested rights. It sends a clear message: conduct that may have been unlawful when undertaken will be treated as lawful because it is now economically indispensable to the broligarchs.

That is not how the rule of law is supposed to work.

Separation of Powers by Suggestion

The framework’s treatment of Congress is equally striking. It does not say Congress lacks authority to legislate. The President cannot say that. Well…he can, but there’s no foundation for the statement. The Constitution is clear: Congress defines copyright.

Instead, the framework says Congress should not act in ways that would affect judicial resolution of the training question.

That is an unusual formulation. Congress legislates in areas under litigation all the time. Indeed, it is often expected to clarify statutory ambiguity.

What the framework is doing is more subtle: It is attempting to shape the legislative field without formally constraining it.

And it pairs that with an implicit second message:

  • Legislation that restricts training or mandates licensing is inconsistent with executive policy.
  • Such legislation is therefore unlikely to be signed by the President. So why bring it?

That is a veto signal—delivered without the political cost of an actual veto.

Judicial Signaling Without Command

The same dynamic applies to the courts.

The framework claims to “defer” to the judiciary. But it simultaneously declares a preferred outcome: training is lawful.

That is not deference. That is signaling.

Judges are, of course, independent. But they do not operate in a vacuum. They are aware of executive priorities, legislative inaction, and market realities. When all three align around a single policy direction, it creates an interpretive gravitational force that is difficult to ignore.

And the signal travels further.

To lawyers.
To regulators.
To anyone whose career may intersect with executive appointment.

It normalizes what counts as a “reasonable” position within the current policy environment.

Prosecutorial Silence as Policy

There is also a more immediate, practical consequence.

While the framework does not have the force of law, it functions as an indirect directive to the Department of Justice. By declaring training lawful as a matter of policy, it signals that federal enforcement resources should not be used to pursue cases premised on the opposite view.

In effect, it tells prosecutors:

Do not spend time considering criminal enforcement for large-scale copyright violations tied to AI training. Do not spend time considering antitrust enforcement against the broligarchs. In fact, don’t spend any time prosecuting anyone regarding AI.

That matters because, for example, willful copyright infringement at scale can, in certain circumstances, give rise to criminal liability. I mean if that doesn’t, what does? Yet under this framework, even the possibility of such enforcement is quietly set aside.

This is not formal immunity. But in practice, it can look very similar.

Why “Not an Executive Order” Matters

If this were an executive order, all of these issues would be front and center:

  • Is this a taking?
  • Does it exceed executive authority?
  • Does it interfere with Congress?
  • Does it interfere with the Judiciary?

Because it is not and EO, these important issues remain in the background—present but untested.

That is the genius, and the danger, of the approach.

It allows the executive branch to:

  • Shape doctrine
  • Influence courts
  • Constrain Congress
  • Guide enforcement priorities
  • Normalize contested conduct

—all without triggering the mechanisms designed to check it.

The Constitutional Shadow

The AI framework does not violate the Constitution in any formal sense.

It does something more complicated.

It operates in the constitutional shadow—where policy can reshape rights, incentives, and expectations without ever crossing the line that would allow a court to say no.

But shadows matter.

Because by the time the law catches up—if it ever does—the world the Constitution was meant to govern and protect may already have changed.

Since it’s 1999, What MGM v. Grokster Teaches Us About Perplexity’s Bizarre Infringement Defense

Nate Garhart writing in Reuters analyzes Perplexity AI’s novel—some might say bizarre—legal defense in copyright suits filed by the New York Times and the Chicago Tribune in December 2025.   Rather than relying primarily on fair use, the typical defense in AI infringement cases, Perplexity instead argues it lacked “volitional conduct” sufficient for direct copyright infringement, contending that it did not “make” the infringing copies in a legally relevant sense. The defense in Perplexity’s motion to dismiss draws on the Second Circuit’s 2008 Cartoon Network v. CSC Holdings decision, where a DVR service was not held directly liable because the user, not the service, initiated the recording of each specific work.  Sound familiar?  That’s one straight outta 1999.  You know, the technology made me do it.

Perplexity-Reply-in-supp-MTD-Feb-27-2026Download

The article explains the strategic logic: eliminating direct infringement would be meaningful, even if secondary liability theories survive.  However, Mr. Garhart is correctly skeptical the defense will succeed, at least at the motion to dismiss stage.  A key difficulty for Perplexity is that Perplexity’s system involves a far more complicated causal chain than a mere DVR: it crawls, scrapes and copies paywalled articles, indexes them, stores them, and generates output that may track the original expression—each step reflecting deliberate system design. Hold that thought, we’ll come back to that one shortly. The newspapers’ attorney, Steven Lieberman, publicly emphasized that Perplexity “does not dispute copying The Times’s journalism from behind a paywall to deliver responses to their customers in real time”.  Rut roh.

The article also flags that the 2nd Circuit may be reluctant to extend Cartoon Network to generative systems that actively synthesize and reproduce expression, rather than merely storing or transmitting content like a DVR.   Ya think?  Ultimately, Mr. Garhart suggests the case may hinge more on fair use—and particularly the fourth factor of market harm that was so attractive to Judge Chhabria in the Kadrey case—because Perplexity’s output product arguably substitutes directly for paywalled news content and floods the market with infringing copies.

Why Generative AI Is Not a Passive Conduit

Mr. Garhart makes clear that Perplexity’s attempt to cast itself as a mere automated tool triggered by user prompts is fundamentally at odds with how generative AI systems actually work. There are several reasons why the “passive conduit” framing fails.

Deliberate System Architecture Embodies Volition

Unlike the DVR in Cartoon Network, where a user pressed “record” to copy a specific program they selected, Perplexity’s system was engineered to crawl restricted content, scrape it and retain it in usable form, and generate responses that substitute for the original.   Mr. Garhart argues that “a company that builds a machine capable of extracting and reproducing paywalled journalism has made many volitional choices, even if those decisions occurred at the design stage rather than at the moment of each query”.  The system’s architecture—its crawling, scraping, indexing, storage, and generation workflow pipeline—is itself the volitional act, not the functionality of a neutral pass-through. 

The Grokster Inducement Framework Reinforces This Analysis

So if we’re going to reach back to 1999 for the defenses, let’s go back for the law, too.  Although Mr. Garhart did not mention it, it’s worth noting that the U.S. Supreme Court’s unanimous 2005 decision in MGM Studios, Inc. v. Grokster, Ltd. held that “one who distributes a device with the object of promoting its use to infringe copyright, as shown by clear expression or other affirmative steps taken to foster infringement, is liable for the resulting acts of infringement by third parties”.  And therein lies the rub.  

The Court identified three particularly notable features of intent evidence:

  1. Targeting a known demand for infringement: Grokster and StreamCast aimed to capture former Napster users, a market defined by its appetite for pirated content. 
  2. Failure to implement filtering or safeguards: Neither defendant developed tools to diminish infringing activity, which—while not independently sufficient—was probative of intent alongside other evidence. 
  3. Revenue model dependent on infringement: The defendants’ advertising-based business model relied on high-volume use, which was overwhelmingly infringing. 

While the Grokster inducement standard requires evidence of purposeful, culpable expression and conduct rather than mere knowledge of infringing potential, the framework’s emphasis on design choices and affirmative steps is directly relevant to generative AI systems like Perplexity. 

Moreover, at each stage of Perplexity’s training pipeline, human decision-making is deeply embedded: engineers and researchers decide what content to tokenize, how to structure training data, and which model behaviors to reinforce or suppress through “reinforcement learning from human feedback” (RLHF) and other fine-tuning methods. The resulting system is curated by humans at multiple points in the typical workflow from dataset selection and preprocessing, to model alignment and quality control, meaning the outputs are not the product of a purely autonomous process but rather of layered, intentional design choices made by people, or more precisely, by Perplexity.

Tokenization itself is a telling example of design choice: by selecting a tokenization scheme and deciding which corpora to process (and spend scarce compute resources on), the system’s developers are making both editorial and commercial judgments about what material the model will learn from and be capable of reproducing. These upstream human choices further undercut the notion that the system is a passive conduit simply responding to downstream user prompts.

Importantly, these tokenization decisions are not made in a vacuum or for altruistic reasons—they are driven by the commercial imperative of delivering a product sufficiently useful that consumers will pay Perplexity for it, rather than paying the New York Times or other original publishers for their journalism. The economic logic is plain: the more effectively the system can ingest and repackage high-quality copyrighted content, the more valuable the product becomes to subscribers, and the more extracted revenue flows to Perplexity instead of to the creators whose work fuels the system. These upstream human choices further undercut the notion that the system is a passive conduit simply responding to user prompts.  Sound familiar?

Applying Grokster‘s Logic to Generative AI

Several design features of a generative AI answer engine map onto the Grokster framework, even without identical facts:

The Causal Chain Is Not Broken by a User Prompt

Unlike the DVR in Cartoon Network where a user pressed “record” to copy a specific program they selected, Perplexity’s system was engineered to crawl restricted content, scrape it and retain it in usable form, and generate responses that substitute for the original. The article argues that “a company that builds a machine capable of extracting and reproducing paywalled journalism has made many [, many] volitional choices, even if those decisions occurred at the design stage rather than at the moment of each query.” The system’s architecture—its crawling, scraping, indexing, storage, and generation pipeline—is itself the volitional act, not a neutral pass-through.

I think Mr. Garhart’s most compelling point is that a user’s query is not the kind of discrete, volitional act that broke the causal chain in Cartoon Network.  A user who types “What does the New York Times say about X?” is asking a question—not selecting a specific copyrighted work and pressing “copy” as with a DVR. The Perplexity system then selects, processes, and generates expressive content drawn from copyrighted sources because that’s how it was trained.   The Grokster Court rejected the notion that intermediaries like Perplexity could hide behind user-initiated actions when those intermediaries had built systems designed to facilitate infringement and had taken affirmative steps to encourage it. 

Critically, the generative AI system’s response to a prompt is shaped by decisions made long before the user ever typed a query. Humans selected the training corpora, decided how text would be tokenized and encoded, fine-tuned the model’s outputs through iterative RLHF and other quality-control processes, and designed the retrieval and generation architecture. Each of these steps reflects purposeful human conduct—not the behavior of a neutral pipe. A system in which humans curate the inputs, architect the processing, and refine the outputs at multiple stages is, by any reasonable measure, an active participant in producing the allegedly infringing content.

In sum, generative AI systems are not passive conduits. They are purpose-built products whose design choices—what to crawl, what to tokenize, how to store it, when to reproduce it, and how to monetize it—reflect exactly the kind of upstream volition and deliberate architecture that both the Cartoon Network volitional conduct doctrine and the Grokster inducement framework are designed to capture. The fact that a user prompt triggers the final output does not absolve a company that engineered every step in the chain leading to that output.

Why did Perplexity scrape leading newspapers for content to feed their AI?  Because it was high value, well written, well editing writing and it was valuable to them.  In short, they did it for the money.

They robbed the authors for the same famous reason Willie Sutton robbed the banks.  Because that’s where the money is.

And going back to 1999 won’t save them.

Sony’s AI Music Attribution Tool: What It Actually Does (and What It Doesn’t)

As generative music systems like Suno and Udio move into the center of copyright debates, one question keeps coming up: Can we actually tell which songs influenced an AI-generated track? And then can we use that determination in a host of other processes like royalty payments?

Recently a number of people have pointed to research from Sony AI as evidence that the answer might be yes. Sony has publicly discussed work on tools designed to analyze the relationship between training data and AI-generated music outputs.

But the reality is a little more nuanced. Sony’s work is interesting and potentially important—but it is often misunderstood. What Sony has described is not a magic detector that can listen to a generated song and instantly reveal every recording the model trained on.

Instead, Sony is describing something more modest—and in some ways more useful.

Let’s unpack what the technology appears to do right now.

Two Problems Sony Is Trying to Solve

Sony AI has publicly discussed research in two related areas.

The first is training-data attribution. This means trying to estimate which recordings in a model’s training dataset influenced a generated output.

The second is musical similarity or version matching. This involves detecting when two pieces of music share meaningful musical material even if they are not exact copies of each other.

Sony has framed both efforts as research directions rather than a finished commercial product. In other words, this is still a developing technical approach, not a turnkey system that can produce definitive copyright answers.

Training Data Attribution in Plain English

The most relevant Sony work is a research project titled Large-Scale Training Data Attribution for Music Generative Models via Unlearning.

That title sounds intimidating, but the basic idea is fairly intuitive and also suggests the project is part of the broader machine unlearning academic discipline.

The system does not operate like Shazam. It does not simply listen to an AI-generated song and say:

“This track was trained on Song X, Song Y, and Song Z.”

Instead, the approach works more like this.

Imagine you already know—or at least suspect—which recordings were used to train the model. You have a candidate set of training tracks.

The system then asks:

Among these training recordings, which ones seem most likely to have influenced this generated output?

In other words, the system ranks influence among known candidates.

The research approach borrows from an area of machine learning called machine unlearning, which studies how particular training examples affect a model’s behavior. In simplified terms, researchers can test how the model behaves when certain training examples are removed or adjusted. If the output changes meaningfully, that suggests those examples had measurable influence.

The important point is that this is an influence-ranking tool, not a forensic detector.

It tries to answer:

“Which of these known training tracks mattered most?”

Not:

“Tell me every song the model was trained on.”

Sony’s Other Idea: Smarter Music Comparison

Sony has also described work on musical similarity detection.

Traditional audio fingerprinting systems—like those used by Shazam or Audible Magic—are very good at identifying identical recordings. If you upload the same song or a slightly altered version, the system can match it.

But generative AI raises a different problem. An AI output might resemble a song musically without copying the recording itself.

Sony’s research tries to detect those kinds of relationships.

For example, a system might notice that two tracks share melodic fragments, rhythmic patterns, harmonic progressions, or musical phrases even if the arrangement, production, or instrumentation is different.

In plain English, this kind of tool tries to answer a different question:

“Are these two pieces of music related in substance?”

Not:

“Are they the exact same recording?”

The Big Limitation: You Still Need the Training Dataset

Here’s the key limitation that often gets overlooked.

Sony’s attribution approach appears to depend on having access to the candidate training dataset.

The system works by comparing a generated output against recordings that are already known or suspected to have been used during training. It estimates influence among those candidates.

That means the system answers the question:

“Which of these training tracks influenced the output?”

But it does not answer the question:

“What unknown recordings were used to train this model?”

If the training corpus is hidden or undisclosed, the attribution system has nothing to test against.

This makes the technology conceptually similar to many machine-learning research experiments, which measure influence using known datasets. Researchers can test influence among known training examples, but they cannot reconstruct an unknown dataset from outputs alone.

What This Could Look Like in the Real World

If the training corpus were known, a practical workflow might look like this.

First, the recordings in the training corpus would be identified. Audio fingerprinting systems could match those recordings to commercial releases.

That step answers the question:

What copyrighted recordings appear in the training data?

Then an attribution tool like the one Sony describes could be used to analyze generated outputs and estimate which of those known recordings appear to have influenced them.

This would not prove copying in every case. But it could dramatically narrow the analysis—from millions of possible influences to a smaller list of likely candidates.

What Sony Has Not Claimed

Sony’s public statements do not suggest that the attribution problem is solved.

Sony has not announced a system that automatically calculates track-by-track royalty payments for AI-generated songs. Nor has it described a tool that conclusively proves copyright copying from an AI output alone.

Instead, the work is framed as research aimed at improving transparency and accountability in generative music systems.

Why Labels Might Still Be Interested

Even with these limitations, the idea could be attractive to rights holders.

If training datasets were known, attribution tools could theoretically support new ways of analyzing how music catalogs interact with generative AI systems.

For example, such tools might help support:

  • royalty allocation models
  • influence-weighted compensation frameworks
  • catalog analytics
  • AI audit trails showing how repertoire contributes to model behavior

In other words, the technology could potentially become a measurement tool for how music catalogs influence generative systems.

What Sony did and did not do (yet)

Sony’s work does not magically reveal every song an AI model trained on. And it does not eliminate the need to know what is in the training dataset.

Instead, its value appears to lie after the training data is known.

Once you have a candidate training corpus, tools like the ones Sony describes may help analyze which recordings influenced particular outputs.

That makes the technology best understood as a post-disclosure attribution layer, not a substitute for knowing what recordings were used in training in the first place.