ZDNET’s key takeaways
- The government is backing OpenAI in major AI lawsuits.
- Ziff Davis (ZDNET’s owner) says creators deserve licensing payments.
- AI needs human content, but may be destroying it.
Ever since generative AI arrived in early 2023, we’ve seen that its almost unlimited base of knowledge is due to how the big AI companies trained their models. To a large degree, AI models like those from OpenAI and Anthropic have been trained on anything they could ingest, including nearly all the copyrighted material on the web.
The AI companies are even reported to be buying up physical books by the millions, cutting them apart, scanning them in, and then disposing of the remains. For example, based on a search of the Anthropic settlement database, I know the company scanned my book, The Flexible Enterprise, and included it in the Claude corpus. I was never asked for permission. I don’t even get a free Claude account.
Due to the bulk ingestion of intellectual property, many companies filed suit against the AI companies. One such company is Ziff Davis, the owner of ZDNET. Disclosure: Ziff is also the company that pays me each week for my writing here.
Two weeks ago, on Sept. 1, 2026, the US Department of Justice filed a Statement of Interest with the US District Court for the Southern District of New York, where the case is being litigated. What makes this statement particularly interesting is that the DOJ is not one of the parties in the case. The government is putting its thumb on the scale, weighing in on a lawsuit between private parties.
Also: Human-in-the-loop oversight is critical for enterprise AI: 4 experts explain why
On Monday, Fortune published a commentary by Vivek Shah, CEO of Ziff Davis, regarding the Justice Department’s unusual intervention in the case. In this article, I’ll briefly summarize the DOJ’s statement, then discuss Shah’s premise, and then pick up and expand upon it with some of my own thoughts.
DOJ Statement of Interest
The Statement of Interest wasn’t solely focused on Ziff’s lawsuit. Instead, it provided government commentary on a broader group of related AI lawsuits, with plaintiffs including The New York Times, the Chicago Tribune, The Sun-Sentinel Company, The San Jose Mercury News, Ziff Davis, and even authors and entertainers like George RR Martin, Michael Chabon, Sarah Silverman, and more.
Essentially, the document is a statement by the US government to the courts overseeing the various related cases, saying, “This is how we think you should rule, and here’s why.”
The government’s position is that the courts should rule in favor of OpenAI and against all the publishers and authors who are trying to protect their rights and the value of their hard work.
Also: Sick of AI slop? How I forced Google to give me results I can trust again
Writing on behalf of the US government, DOJ Associate Attorney General Stanley E Woodward Jr., Assistant Attorney General for the Civil Division Brett Shumate, and Senior Counsel to the Associate Attorney General Michael Weisbuch make the claim that AI training should be classified as fair use.
Their premise is that because, “Training of LLMS on written works is exceedingly transformative,” the new work, “adds something new, with a further purpose or different character, altering the first with new expression, meaning, or message.”
Fair use is the legal doctrine that copyrighted material can be used without permission in certain circumstances that encourage creativity, commentary, and learning. As you might imagine, even way before AI, copyright owners and those who have wanted to expand on others’ IP (intellectual property) have disputed the various interpretations of what constitutes fair use.
The DOJ makes four other key points in its statement.
1. The DOJ rejects the idea that AI-generated works in the same genre as the original can cause market harm. Any website operator who’s been trying to run a site since Google introduced AI Overviews can make a very strong case that the DOJ is smoking something here. On average, sites have lost almost 40% of their search engine traffic due to AI Overviews.
2. The DOJ claims that limiting AI training could hinder US AI development. Here, they have a point. But we generally don’t allow stealing in this country. This is a tough line, because the competitiveness and national security concerns of the government are real and valid.
But sucking down my comparative review of vertical mice and then regurgitating it to ChatGPT users will do nothing to increase either our national security or our competitiveness.
3. Requiring licenses to content could raise the barriers to entry for AI companies to produce LLMs. This is just downright idiotic. The barrier to entry for a new LLM is compute power on a vast, almost planetary scale.
Look, Spotify and a bunch of other music services license music (including my EPs), and the pittance they pay artists certainly doesn’t hinder their development. Sure, adding licensing fees will add cost. But, as I’ll discuss in a bit, the societal cost of putting authors and publishers out of business is vastly higher, even to the AI companies themselves.
4. Foreign competitors will steal copyrighted content anyway. This is the premise that LLM makers outside the US won’t respect copyright, so to give our own models an equivalent advantage, we should ignore it as well.
The fact is, foreign entities have been ignoring US copyrights forever. It’s not a new thing that just spawned in the AI era. But we still protect creators’ rights in this country. Just because other nations engage in predatory behavior against US citizens, that doesn’t mean we should as well. Both the big publishers and indie authors have constant disputes with infringers, but at least the US portion of the business has essential protections.
The DOJ is advocating throwing that out for the AI companies.
Ziff Davis’ response
CEO Shah’s headline premise is that “If OpenAI can find $750 billion for data centers, it can find money for publishers, too.”
He has a point. A billion here. A billion there. Pretty soon, you’re talking real money.
Shah takes issue with the government’s claim that licensing costs could make the creation of AI models too expensive and raise barriers of entry to AI model developers. His counter is, “There’s no effort to estimate the capital required for licensing or compare it to what’s being spent on data centers, chips, and power. OpenAI alone has told investors it expects to spend $750 billion on compute by 2030.”
He adds, “For context, royalties and licensing payments in the US music industry are under $20 billion per year.” His premise is that if similar amounts were paid to publishers, the money would be virtually unnoticeable to the large AI companies, but would be “meaningful to content producers of all sizes.”
I believe he’s completely correct when he critiques the Justice Department’s premise that AI-generated works don’t cause market harm. Shah contends that unlicensed content use is an existential threat to the economics of producing original content.
He also takes on the Justice Department’s premise that requiring content licenses would create friction when it comes to developing US-based AI systems. Shah’s contention is that, “In the past 18 months, a number of services have been launched to make licensing even more turnkey, streamlined, and efficient.”
Also: ‘Sophisticated’ AI swarm attacks are months away, OpenAI warns: What businesses must do
He cites the example of ASCAP (American Society of Composers, Authors and Publishers) and BMI (Broadcast Music, Inc.). Both of these are performance rights organizations that arrange for royalties for creators when their works are used according to certain conditions.
I’m registered with BMI, for example. That means, if my music is used commercially, there’s a structure in place where I automatically get compensated. At my level of music success, it’s pennies. But at least it’s something. It’s certainly more fair than the big companies just taking my work without any compensation or permission.
When it comes to whether foreign adversaries (or even frenemies) respect American licensing and copyright requirements, Shah doesn’t fight the premise. Instead, he says, “Likely true. But do we want to abandon our respect for intellectual property just because our adversaries have?”
That’s when he dives back into the cost comparisons between building billion-dollar data centers and licensing content. Sure, doing what AIs do is fantastically expensive, but he contends that is “why there are only a handful of large-language model companies in the world with access to the capital required to create frontier models. The barrier to entry here has nothing to do with publisher licenses.”
The cost, he says, is that, “Without a licensing framework, publishers will increasingly erect barriers and paywalls to restrict their own content and extract those fees. This imposes substantial costs to consumers, limits access to free information and content on the open web, and reduces pathways to earnings for writers and content producers.”
There is a national security concern
If we create a world where content producers can’t make a living, we run the risk of stagnating the internet information pool. What would the world be like if no new content was produced after 2023?
Anyone who’s seen Idiocracy has a fairly vivid picture of what the world would be without good, original, well-considered content.
In the movie, Luke Wilson plays a corporal who was selected for a top-secret hibernation program. They forget about him, and he sleeps for 500 years. When he wakes up, it’s to a future that’s been thoroughly dumbed down and is ideologically insular to the point that his common level of intelligence is considered genius.
That’s not the world I want to create or support. That’s why Shah’s comments are so important, beyond the mere economics of a tech publisher. In this regard, I’d like to share five thoughts that came to mind when considering his editorial.
1. A healthy content economy is necessary, both for AI and for society as a whole. Think about companies like OpenAI and Google. Both would have almost no value if they couldn’t leverage content produced by everyone else.
For Google, at least initially, without something to search for, there was no reason for a search engine. Without its huge corpus of general knowledge, OpenAI’s ChatGPT wouldn’t have been able to engage in conversations. Back in the 1980s and 1990s, we had AI without ChatGPT-sized data stores. We called them expert systems, and they consisted of the carefully curated knowledge of one or two very well-defined domains of knowledge.
Do you remember the huge AI boom of the late 1980s and 1990s? No? That’s because it didn’t happen. Back when I created my AI products during those periods, when I approached investors, as soon as they heard the acronym AI, their eyes glazed over.
More than one told me, “There is absolutely no market in AI.” And, you know what? Without the entire, vast content economy we’ve created fueling it, and without the massive data centers able to store and regurgitate that knowledge, and without ubiquitous broadband, there would be no real money in AI.
So what happens when AI puts content developers out of business?
Take this article, for example. I wouldn’t be able to spend the hours it takes to research and write this if I weren’t being compensated. I’d need to put that time into finding another way to put food on the table for my family. Multiply that by all content developers, and you lose the entire band of professional producers.
All that would be left are producers of AI slop, those who produce content to fulfill an agenda, those who produce for the fun of it, and those who produce content to corrupt existing content. Without professional content producers supported in their work, we would slide toward Idiocracy.
2. The open-web bargain is being destroyed. The bargain that all of us web producers made with Google was a simple one: you can index our content. In return, you’ll send us traffic that allows us to derive value from our work.
Although there are paywalls and subscription programs, the growth of the web has been fueled primarily by free-to-read content. Sites like ZDNET have been ad-supported or supported by other commerce programs that allowed people like me to get paid to create content and allowed people like you to read it without first forking over credit card digits.
But while we all tacitly agreed to free-to-read, we never agreed to free-to-industrialize at an enormous scale. Search engines indexed our content but returned traffic. AIs, including Google’s AI Overviews, are using our content and holding our readers captive.
I see my friends doing this all the time. They’ll type something into Google, get an AI Overview from Gemini, maybe ask Gemini a few questions, and leave the experience satisfied. Rather than click through to read an article, they have their answers immediately.
Yes, that does add to reader convenience. But there will come a time when those answers become stale if those who write the articles the answers are derived from are no longer able to make a living from the practice.
That free-to-read economic model will break down, everything will be fueled by subscriptions, crawlers will be blocked, and we’ll all have much less access to the vast troves of information we’ve become used to having at our fingertips.
3. Licensing is not that much of a technical obstacle. It’s just not. Licensing, in various forms, has existed nearly forever. Digital licensing and microtransactions are merely a matter of some API calls. For the US government to claim that licensing alone will diminish American competitiveness is simply unsupportable.
For that matter, as Shah said earlier, licensing isn’t really even an economic obstacle for the AI companies. He called the amount that licensing would cost, compared to the cost of data centers, a “rounding error.” I agree.
4. Licensing needs to work universally. Look, I know that Taylor Swift gets a whole lot more plays on Spotify than I do, but we’re both verified by Spotify as artists, and we’re both affiliated with BMI for performance rights licensing. She has roughly 100 million more monthly listeners than I do. Yet the same licensing service is available to us both.
Yes, signing up was an annoyance. And, yes, it took some time to figure out what I was doing. But I didn’t need to have millions of fans to be able to set up my music licensing. So why should this be considered such a huge, national security-threatening barrier for other content?
Answer: it’s not. It just needs to be available to all content producers and consumers. It needs to be auditable. And it needs to work.
Here’s one thing to consider: training, real-time retrieval, quotation, summarization, and answer generation may need to have different classes of licensing.
5. What information ecosystem do we really want? I’ve spoken over and over about how AI is fantastic and stupid, all at the same time. I’ll admit that there are days I wish generative AI never happened. But that’s mostly on days when I’m worried that, as a content creator, AI and corporate cost-cutting could reduce my ability to make a living.
On other days, I can see the enormous value in AI. Just this summer, I wrote a couple of articles about how I derive fairly tremendous value from my $20-per-month ChatGPT account and ChatGPT Work, the agentic tool built into that subscription.
Also: 5 ways I actually use ChatGPT Work in my small business to save time and money
To a degree, agentic AI turns the information exchange on its head. Now, not only are humans reading websites, so are AI agents. Fortunately or unfortunately, an AI agent is unlikely to be tempted by an ad while reading about, for example, how software products implement undo chains (this is a task I gave Claude Code agents just this morning).
We need fresh information and content
For AI to function and for our world to grow, we need fresh information and fresh content. I remember a time when all information was gated behind paywalls. Physical ones. I had a small hometown library. If I wanted to know anything else, I had to buy the book. But even if I had the money to buy the book I wanted, the bookstores were so small that the sort of arcane topics I was interested in were not stocked in the mall bookstore. Maybe they could order a book for me from their suppliers. Maybe they couldn’t. I am sure there were books I would have been interested in, but I never even knew those books existed, and I had no way to find out.
Amazon made mail-order books easier to access. Then came the web. It had an impact. I used to have five to 10 specialty magazine subscriptions (and stacks of those magazines in the corner of my living room). But once I could read online, I stopped bringing all that paper into the house.
Yes, I still pay for some subscriptions, but they’re all online now. Online has disrupted everything from periodicals to classified ads to taxi services. We have to be able to absorb disruption and change. But to do that, we have to remain fresh, aware, and connected.
Government policy must not favor one information industry over another. I challenge the DOJ with this question: if you think AI competitiveness is essential to national security, what happens when our entire country stops innovating because creating new knowledge is no longer economically viable?
Also: The best and worst AI for your privacy, ranked – and how each handles your data
If we let AI break the economic structure that supports creating new content, and that content was what fueled the open web, what are we going to replace it with? Would you prefer Americans read RT.com instead of The New York Times? RT stands for Russia Today, and it’s part of ANO TV-Novosti, “The Russian government entity responsible for the worldwide broadcasts,” of the Russian propaganda machine.
RT, for example, syndicates much of its content through social media like Facebook. I’ve seen many Facebook re-posts of RT material by Americans who don’t realize they’re reposting Russian state propaganda. If news organizations and editorial outlets can’t be profitable because of AI scraping, then where will fresh new content come from? If editorial profitability is out of the picture, the new-content money will come from vendors selling stuff, state-run organizations, and those with an agenda they want to foist on readers and push into AI algorithms.
State-run propaganda outlets or news outlets will probably be considered authoritative to both search engines and AI training sources. If those organizations that commerce and traffic is currently funding, like The New York Times or Ziff, aren’t able to make their expenses, then state-supported news organizations will have a clear winning opportunity to be the new sources of truth.
I would contend that the internet, for the past 30 years, has not always had accurate information or vetted information, but it’s been vibrant with human information, human thoughts, human ramblings, and such. It’s not necessarily that the content produced needs to be something purely accurate. It’s that it needs the full breadth of human input, and not just content produced by those who are supported by state media or hype hucksters.
But if we allow AI to put American news and professional content organizations out of business, our frenemies will happily take their place, producing vast pools of new content that is most definitely not in our national interest. Basically, if our professional content producers are pushed out due to scraping, and the door is left open for those funded by nation-state interests, our entire information flow either stagnates or becomes a national security risk, or both.
I’m not agreeing with Shah’s arguments because he’s somewhere up in the stratosphere of ZDNET’s management structure. I agree with him because I make my living producing content, sharing my years of experience, and working very hard to provide you with value.
Plus, I, like almost everybody else on the planet, have truly benefited from the expansion of knowledge on the internet, but I also have a national security perspective that I can bring to bear due to my work in counterterrorism, infrastructure security, and cybersecurity. I can look at these things from both points of view. Either way, it’s very important to sustain the free market of information on the internet because I can see the truly big risk that will come from allowing only vested interests and not a free market to produce the information that we consume.
And, with that, here’s another thought from me to the Justice Department.
If, as the government’s Statement of Interest claims, AI is truly a national security priority, then nurturing and preserving the economic viability of the creators and companies who create the content that fuels AI must also be considered a national security priority.
Do you think AI training on copyrighted work is fair use, or should creators be paid for it? Let us know in the comments below.
You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.
