🌐 AI splinternet

🏛️ Scoop: DOJ's filing blindsides agencies

📈 Deals leapfrog suits

🕸️ New entrants tangle AI-media web

⛓️‍💥 Link economy breaks

🪪 Deepfakes drive AI state laws

🎬 Hollywood plays nice

🎼 AI music drives legal frenzy

📚 AI supercharges audiobook piracy

AI is scrambling decades of global copyright and fair-use precedent, presenting new challenges for content distributed across borders.

Why it matters: Large language models can train on content in countries where that use is legal and then be deployed in countries where it's not.

Zoom in: Open models are particularly difficult to contain because they can be downloaded, modified and redeployed.

Open Chinese AI models have already been downloaded more than 10 billion times, Vermillio CEO Dan Neely says, likely spawning many times that number of customized versions built for specific uses.

🌏 Zoom out: This dynamic could force global platforms to treat the exact same AI-generated content differently depending on where a user lives.

"It's going to be a nightmare to handle," Deezer vice president of research Manuel Moussallam says. "The big question is who's going to audit this model and determine that the rules are OK in one territory and not the same in a different one."

State of play: New licensing and permissions infrastructure is already being designed to address that question.

Really Simple Licensing (RSL), a machine-readable open standard, lets copyright holders specify how AI platforms can use their content and set geographic parameters, co-founder Doug Leeds told Axios. 

Companies like Snowflake and Cloudflare are also offering solutions at the cloud-serving level.

Context: Fights around fair use mainly center on two debates: one, whether AI firms can legally use copyrighted works to train their large language models, and two, whether they can reproduce or closely resemble those works in the outputs.

The Justice Department recently indicated it believes using copyrighted works to train models should be considered fair use, but outputs aren't necessarily covered by the same legal principles.

The DOJ's position is not binding, but if a court adopts its reasoning, that could make it harder for copyright holders to enforce training-based infringement claims under U.S. law, including those against foreign AI firms.

🎞️ Between the lines: Hollywood's fight with China's MiniMax illustrates how studios are trying to leverage U.S. copyright protections to target AI models developed overseas, but whose outputs can be used in the U.S.

Getty Images' litigation against Stability AI shows how copyright protections can stop at borders, even when models are distributed beyond them.

What to watch: Some IP holders are trying different legal tactics to circumvent trickier fair-use cases.

1 big thing: 🌐 AI splinternet

Illustration of a smartphone being cut in three by two lines of binary code.

Illustration: Brendan Lynch/Axios. Stock: Getty Images

The Department of Justice's statement of interest supporting OpenAI and Microsoft in the New York Times' copyright infringement lawsuit surprised critical agencies like the U.S. Patent and Trademark Office and the Copyright Office, sources told Axios.

Why it matters: Statements of interest allow the government to declare an official position on a legal matter in private lawsuits. While not binding, they can hold significant weight and help persuade cases.

⚖️ Zoom in: The Times' lawsuit against OpenAI and Microsoft argues AI firms illegally stole their copyrighted works to train their large language models, and that AI-generated outputs unlawfully produced market substitutions.

The DOJ's SOI argues copyrighted works to train models should be considered fair use because that practice is new and transformative, but also says outputs aren't necessarily covered by that same legal argument.

Unlike many SOIs, no career antitrust attorneys signed the filing alongside senior DOJ officials.

Between the lines: Publishers have criticized the claims in the SOI, including the idea that enforcing copyright laws is too cumbersome and would threaten America's AI dominance over foreign rivals.

Vivek Shah, CEO of publicly traded publishing behemoth Ziff Davis, pointed out in a recent op-ed that the government's argument is muddled. 

The DOJ argues in its brief that licensing enforcement is cumbersome, but doesn't take a position on whether a licensing regime is feasible.

Reality check: The fact that the government's position on this issue was not coordinated among the bodies that work closely on it, especially the Copyright Office, is notable, but not shocking.

President Trump abruptly fired the Copyright Office's longtime director, Shira Perlmutter, last year. Perlmutter sued to block the move, and the legal process is ongoing.

Last year, the DOJ settled an antitrust lawsuit brought by the Biden administration that sought to break up Live Nation and Ticketmaster, despite bipartisan backing for the case.

Shortly after that decision, made by a small group of DOJ lawyers, the Wall Street Journal reported that career attorneys working on the case learned about the settlement when the court did.

The U.S. Patent and Trademark Office and the Copyright Office declined to comment. The Justice Department didn't respond to a request for comment.

2. 🏛️ Scoop: DOJ's filing blindsides agencies

Illustration of the Scales of Justice, with a line of binary code pushing down one of the sides.

Data: Axios research; Chart: Erin Davis/Axios Visuals

Deals with AI firms have dramatically outpaced lawsuits over the past two years, according to an Axios analysis, suggesting creative industries are eager to strike commercial agreements amid ongoing litigation.

Why it matters: The licensing market is developing faster than courts are establishing the underlying rules.

Only a handful of cases have produced major decisions so far and many others could spend years working their way through district and appeals courts.

By the numbers: AI companies have signed at least 299 media deals since 2023, outpacing the 107 lawsuits filed over the same period, according to an Axios analysis.

📰 News has moved furthest into the licensing market, although much of that momentum is coming from deals with smaller firms, like ProRata, rather than major companies like Anthropic. 

🎹 Music and book companies have become more litigious over the past two years. The ratio of deals to lawsuits for these companies is much smaller compared to news media, in part because of industry structure.

🎮 The gaming industry has struck few AI deals, although agreements such as Electronic Arts' partnership with Stability AI suggest that could be starting to change.

The intrigue: Deals offer AI firms an insurance policy against further legal action while major cases play out. But litigation has also become an expected part of the negotiation process around licensing.

As part of a settlement between record label giant Warner Music Group and Suno last year, the AI music startup agreed to shut down its original large language model and co-build a new one with record labels on fully licensed material.

What to watch: Dealmaking is shifting from licensing content to train LLMs toward new payment models, including retrieval-based agreements that compensate publishers based on usage.

3. 📈 Deals leapfrog suits

The web of deals and disputes between media and AI companies has ballooned over the past year.

Why it matters: A licensing market once dominated by OpenAI has splintered. Litigation has also dramatically expanded to more companies.

State of play: Lawsuits targeting OpenAI and Microsoft represented around 36% of total AI copyright cases filed before 2025.

That share dropped to 16% of all cases filed since the start of 2025, according to an Axios analysis.

By the numbers: OpenAI now has at least 24 deals with content companies, but rivals have rapidly built their own networks.

The map includes 20 deals for Google, 14 for Meta, 12 for Perplexity, 11 for Microsoft and 10 for Snowflake.

ProRata has built the largest network, with 140 live partnerships spanning 1,635 publications, ranging from large outlets to niche and local publishers.

Between the lines: OpenAI has used its deals as a public display of its commitment to cooperation, while other firms like Google and Anthropic are much more tight-lipped.

The intrigue: Anthropic has yet to emerge as a major partner for news outlets, in part because it relies primarily on enterprise and developer-focused customers.

It has been able to avoid some of the legal scrutiny from digital publishers that OpenAI has faced simply by launching later, Adweek notes. But it's faced a growing number of copyright lawsuits from the music and book industries in the past two years, per Axios' analysis. 

The big picture: Most new lawsuits refer back to precedents set by a handful of cases that have already been litigated, but copyright parameters for the AI era are still very much in flux.

What to watch: A landmark lawsuit against OpenAI and Microsoft brought by the New York Times and other regional newspapers could set a historic precedent around copyright law.

More deals could be brokered once further legal precedent is established.

4. 🕸️ New entrants tangle AI-media web

Data: Similarweb; Chart: Erin Davis/Axios Visuals

Traffic to news sites is plummeting in the U.S. as tech giants shift users from search and social sites to AI interfaces.

Why it matters: That dynamic is forcing digital publishers to pivot faster than many are prepared, driving layoffs and consolidation. Stopgap efforts to slow the decline are costly and unsustainable.

📉 By the numbers: U.S. web traffic to roughly 100 of the top news sites fell to 47.6 million visits in July, down from peaks above 70 million in 2024, according to Similarweb data.

Average monthly search referrals to top publishers fell from roughly 18 million in July 2024 to about 11 million this past July.

AI chatbot referrals have surged from roughly 10,000 in July 2023 to more than 160,000 in July. But that remains a fraction of the traffic traditional search provides.

Context: 2024 marked a traffic peak before the digital tides started turning on publishers.

That was the year Google introduced AI overviews within its search results and Meta rolled out a heavily expanded version of its Meta AI assistant. 

🔍 Between the lines: Publishers are scrambling to make up those losses in the short term by buying search traffic and leaning on affiliate recommendations. But if current trends hold, those efforts will eventually be moot.

In addition to paid search traffic, which has increased 160% since 2023, email and affiliate referrals have been the only growth areas for publishers for the past three years, according to Similarweb data.

Direct traffic, traditionally one of publishers' largest funnels, has sunk 25%. Social media referrals are also down, but publishers are making up that gap by buying more social promotions.

What to watch: Generative AI search engines prioritize organic mentions over paid promotions, creating a new challenge for companies that rely on paid customer acquisition.

5. ⛓️‍💥 Link economy breaks

Data: National Conference of State Legislatures; Chart: Erin Davis/Axios Visuals

States have enacted 128 bills targeting deepfakes, making AI-manipulated content the most legislated AI issue in the U.S., according to an Axios analysis of National Conference of State Legislatures data. 

Why it matters: Deepfakes are among the AI issues that are easier for lawmakers to tackle, because existing laws around impersonation, deception and commercial rights provide a legal framework that can extend to AI.

📜 Zoom in: Most state-level deepfake legislation focuses on transparency measures, requiring disclosures when a person's likeness is used to create new content or when AI is used to create political ads. 

The most prominent state law is Tennessee's 2024 ELVIS (Ensuring Likeness Voice and Image Security) Act, which extended the state's existing personal rights laws to a person's image, voice and likeness. 

Dozens of states have enacted bills that address deepfakes in political advertising and typically require campaigns to label or disclose AI use. 

At the federal level, the No Fakes Act has received wide industry support, but has yet to pass. The measure would establish a federal right law to ensure a person's voice and visual likeness cannot be used for unauthorized digital replicas. 

State of play: Creative industries would prefer a single federal measure that requires consent and transparency around when a person's voice, image or likeness is used in an AI-created work, but a gridlocked Congress means lobbying efforts have shifted to the states.

The big picture: Many state AI bills center on transparency, including whether, when and how that use should be disclosed.

Those efforts seem straightforward, but have been challenging to implement as they require broad consensus over what needs to be disclosed, how often, to whom and why. 

What to watch: States are introducing more bills targeting the private sector's use of AI technology as the country grapples with growing demand for data centers and other AI infrastructure. 

National Conference of State Legislatures

6. 🪪 Deepfakes drive AI state laws

Hollywood is beginning to cut deals that could establish the rules for how studios use AI technology.

Why it matters: The moves come after years of the entertainment industry treating the technology as an existential threat to its business and creative workforce.

🍿 The latest: Disney named Character.ai CEO Karandeep Anand its first chief technology officer yesterday, with members of Character.ai's technical team expected to join him.

The announcement comes about a year after Disney sent Character.ai a cease-and-desist letter accusing the platform of unauthorized use of Disney characters. 

State of play: Studios are moving from just trying to stop unauthorized AI use toward defining the conditions under which it can happen.

The Motion Picture Association reached an agreement with ByteDance over guardrails for its Seedance and Seedream AI tools, following threats of legal action from Hollywood studios.

A24 is working with Google's DeepMind, giving filmmakers access to AI tools and providing feedback to help shape how those tools are developed.

Runway has struck a growing number of Hollywood partnerships, including with Lionsgate and AMC Networks.

📑 Hollywood's labor contracts are getting more specific about the rules for AI.

The Writers Guild's contract requires studios to notify the union if they license writers' work to train a commercial generative AI system.

SAG-AFTRA's contract limits the use of synthetic performers, allowing producers to use them only when they bring "significant additional value" compared with a live performer or that performer's digital replica.

The Directors Guild's contract gives directors control over AI-generated footage and creates an employer-funded program to train members on AI tools.

🎥 What to watch: Filmmakers are looking to AI to solve creative problems that previously would have been too expensive or difficult.

7. 🎬 Hollywood plays nice

Illustration of a movie projector, projecting a line of binary code.

Data: Deezer; Chart: Erin Davis/Axios Visuals

AI-generated music is flooding streaming platforms, but music's unusually complicated web of copyright makes it one of the hardest creative industries for AI companies to navigate legally.

Why it matters: The legal uncertainty isn't slowing adoption and dealmaking.

💿 By the numbers: Fully AI-generated tracks surpassed half of all new music uploaded to Deezer's streaming service at the June peak, up from just 10% in January 2025.

But fully AI-generated music accounts for just 1% to 3% of streams on Deezer, and up to 85% of those streams were identified as fraudulent in 2025.

"People don't really consume AI music," Deezer's Moussallam says, noting exceptions tend to be viral AI songs that listeners seek out knowing they're AI-generated.

Zoom in: Part of what makes music so difficult for AI licensing is that a song isn't protected by a single right or owned by a single party.

Every recorded song contains at least two distinct copyrighted works: the musical composition, typically controlled by songwriters and publishers, and the sound recording, typically controlled by performers and record labels.

Copyright also isn't the end of it. AI products that imitate a performer may also implicate voice and likeness rights that a record label doesn't necessarily control.

🤝 State of play: The music industry is beginning to strike deals. 

Suno last week launched its new v6 models developed with Warner Music Group, BMG and Believe, its first generation of models developed with music-industry partners.

Suno chief product officer Jack Brody says the partnerships establish a revenue-sharing system in which a portion of revenue generated through the partnerships goes to rights holders. 

Brody says Suno is working toward licensed products that let users create derivative works from known artists and catalogs, which its current models don't allow.

Yes, but: The new v6 models are still causing headaches for record labels that haven't settled with Suno.

On Friday, Universal Music and Sony filed a second copyright lawsuit against the company, arguing it trained its new models on their copyrighted works without permission.

8. 🎼 AI music drives legal frenzy

Unauthorized AI-generated audiobooks are proliferating across YouTube and Spotify.

Why it matters: Unlike the debate over training data, AI audiobooks directly compete with what publishers and authors are trying to sell.

🎧 By the numbers: Vermillio identified 18,000 unauthorized AI-generated versions of 1,300 original audiobook titles, or nearly 14 unauthorized versions for every original title, the company shared with Axios.

Those unauthorized versions generated 382 million streams, averaging more than 21,000 streams per piece of content, according to data collected between July 1 and Sept. 1.

Vermillio estimates those streams represented $1.3 billion in value across its sample and extrapolates that to a potential $6.7 billion across an estimated 90,000 audiobooks produced in a year.

Vermillio says its technology scans all major platforms where audiobooks are available online.

🏴‍☠️ Zoom in: AI has "supersized book piracy," Rachel Geman, an attorney at Lieff Cabraser who represented authors in copyright litigation against Anthropic, tells Axios.

The big challenge is enforcement. Rightsholders have to find unauthorized versions scattered across platforms, which Geman likens to a "Where's Waldo" game. 

Audiobooks can be harder to detect than pirated music because copies can be altered, such as by changing words, omitting passages or using an AI narrator, to evade systems designed to identify the original recording.

State of play: Platforms like Spotify and YouTube are adapting their copyright systems to better detect and remove infringing AI-generated audio.

The Audio Publishers Association has hired an outside firm that identified a network of roughly 17 sites distributing audiobook downloads. The group plans to target piracy at those sources, executive director Jim Dinegar tells Axios.

Yes, but: AI could expand the legitimate audiobook market. Fewer than 2 million audiobooks are available globally compared with an estimated 100 million to 150 million print and ebooks, Audible CEO Bob Carrigan said earlier this year.

Tens of thousands of authors have used Audible and Amazon's AI production tools to create hundreds of thousands of audiobooks, Audible tells Axios. Listeners have spent more than 70 million hours with AI-produced titles, which have an average customer rating above four stars.

📖 Zoom out: Books have become one of the key battlegrounds in the fight over how AI companies obtain and use copyrighted material.

Anthropic agreed to settle with authors after a federal judge ruled that training its AI on lawfully acquired books could be fair use but that downloading millions of pirated books infringed copyright.

More than a dozen civil society groups last month urged the FTC to investigate AI companies for buying, scanning and destroying physical books for training, arguing the practice could harm competition.

What to watch: Another major test is underway with the lawsuit against OpenAI and Microsoft from authors including Jodi Picoult and George R.R. Martin.

Axios

Axios Media Trends Executive

September 19, 2026

1 big thing: 🌐 AI splinternet