Brad Pitt Has Figured Out the One Thing AI Can’t Give Hollywood

💪 Fitness Influencer’s Book Hit With AI Accusation, But There’s a Catch

1. Lately, the most popular witch hunt is tracking down books that appear to be AI-generated, and now there’s a new target: The Aesthetic Revolution: Embracing Vanity and the Future of Fitness to Unlock Your Healthiest Self, by professional bodybuilder Mike Israetel, who holds a PhD in sport physiology. The 304-page book was published in June by Grand Central, an imprint of the Hachette Book Group, the second-largest trade publisher globally. The AI accusation comes from Australian YouTuber Solomon Nelson, a self-described natural bodybuilder who posted a surgical, nearly 2-hour-long video dissection of the book on his channel with over 104,000 subscribers.

After going through the book’s topics, chapter by chapter, he then explains the areas he believes to be AI-generated, with fairly convincing arguments. This is the second notable AI accusation Hachette’s authors have faced this year. Back in March, a Hachette book titled Shy Girl, by Mia Ballard, was canceled after a YouTube book reviewer pointed out a number of areas where AI-generated text seemed, according to the reviewer, to have been used in the book.

“If the cancellation of Shy Girl is Hachette’s operational standard for dealing with generative AI, then Hachette’s path forward regarding Mike Israetel’s book is clear,” said Nelson during his video breakdown.

“By any objective metric, the evidence of AI use in The Aesthetic Revolution is far more conspicuous and undeniable than the text-based allegations that brought down Shy Girl. To maintain editorial integrity and consistency, Hachette must apply the same standard to The Aesthetic Revolution. Otherwise, this would signal a double standard. It would prove that Hachette is willing to police AI slop among fiction authors, but turns a blind eye when a prominent fitness influencer cuts the same corners with non-fiction. For Hachette Book Group, the burden of proof has been met. Based on their own track record, the only consistent response is to pull The Aesthetic Revolutionfrom publication.”

🙋🏻‍♂️🤖 MAN AND ROBOTS lands in your inbox every Friday: AI, Hollywood, and the future of work. Get it free →

Disclosure As AI Accusation Antidote: The only reason I found out about this critique of Israetel’s book is that, at various times in my life, I’ve been an amateur non-competing bodybuilder. So I occasionally peek in and see what the latest developments are in the space. I’ve never used steroids, but I understand why some people do. I also understand how attitudes have softened on the topic over the years. Some believe that using steroids is cheating, no matter what. Others believe that steroids are fine, as long as you disclose steroid use before offering the public training advice or presenting yourself as a natural bodybuilder. I’m beginning to see generative AI in the same way.

Yes, I still believe that the makers of most commercial AI models obtained data from humans who created original work, and those humans deserve compensation in some form. However, now that the AI genie is out of the bottle, and people are using it to write books, produce music, and create images and videos, I think the notion that AI will suddenly vanish due to some moral, creative purist, or even legal influence is not realistic. Since AI in the creative arts business isn’t going away, I think the most realistic path forward is to simply demand disclosure regarding when AI has been used. If the audience is informed that AI was used, and still wants to consume that product, then they have some agency and haven’t been deceived.

The problem with a creator using AI and not disclosing it is (at minimum) twofold: 1. A publisher needs to know what it’s attempting to assert copyright over, and 2. The audience deserves to know what they’re actually paying for with their dollars or attention. Similarly, if a bodybuilder like Israetel is on YouTube giving his 3.89 million subscribers workout tips, it’s important that his viewers know that he used steroids to aid his physical appearance. To his credit, Israetel has indeed publicly acknowledgedhis steroid use.

Fittingly, his approach is consistent when it comes to generative AI. Several months ago, before his book was even published, he posted a message on social media outlining his AI use in his then-upcoming book.

“I 100% used AI to help me smooth over some of the prose for my upcoming book, The Aesthetic Revolution. This was before a whole team of editors from the publishing company further smoothed the prose over. They were made aware of this,” wrote Israetel in May. “I’m stating this now, well before the book comes out, so that there is no controversy if and when someone detects that it has elements of AI writing styles in it. I will also be using AI in the future unapologetically to improve my work in various ways.”

As for how U.S. copyright laws will handle this kind of assertion of authorship related to copyright, the U.S. Copyright Office made its position clear back in 2025: “Where a human inputs their own copyrightable work and that work is perceptible in the output, they will be the author of at least that portion of the output. Their own creative expression will be protected by copyright, with a scope analogous to that in a derivative work. Just as derivative work protection is limited to the material added by the later author, copyright in this type of AI-generated output would cover the perceptible human expression. It may also cover the selection, coordination, and arrangement of the human-authored and AI-generated material, even though it would not extend to the AI-generated elements standing alone.”

How major publishers like Hachette will handle this tricky new copyright territory when eventually forced to defend it in the future remains to be seen. During the Ballard incident, the publisher told The New York Times, “Hachette remains committed to protecting original creative expression and storytelling.” Hachette hasn’t commented publicly on Israetel’s AI use, so it could be that the only difference in the two situations was, again, full disclosure before publication. At the very least, until we have more major cases testing the U.S. Copyright Office’s guidelines, I think disclosure is where things must start with these kinds of AI-assisted or AI-generated books.

🎭 Tyler Durden Vs. AI

2. “There’s a lot of pushback against AI, but AI is going to be the very thing that, if used as a tool, is going to help these mid-budget films, forty- to ninety-million-dollar budgeted films, get made… I heard an AI song the other day that was pretty damn good. But it had a repetitive nature to it. It’s not going to be the same as hearing her voice—a woman’s voice, a person’s voice. I saw a show the other week in Paris. I saw Flea and Thom Yorke get up onstage, and they did Marvin Gaye. With Flea’s jazz band. Ohhh, it was off the charts, full of joy. AI can’t do that.”
—Brad Pitt, actor, Once Upon a Time… in Hollywood, Fight Club
*Source: Esquire

🤖 Anthropic’s Plan: Drown Human Work in Watermarks

3. This last bit is more of a warning than anything, and it has to do with Anthropic’s new plan to watermark its Claude output as a way to comply with the European Union’s AI Act, Article 50(2) Code of Practice on Transparency of AI-Generated Content. Specifically, Anthropic says, “Claude models launched in the EU on or after August 2, 2026 will support machine-readable marking at launch. Generated text will carry embedded watermarks, and generated files will include digitally signed provenance metadata where supported.” Sounds good, right? Now, everyone in the EU will know when they’re getting AI outputs rather than directly human-created text or code.

But there’s a problem. The watermarking will not just be affixed to AI-generated text, it will also be attached to any outputs that pass through Claude. That means that if you wrote an original story, with zero AI text, but sent it through Claude for editing and then used that text, your human-created text could be incorrectly watermarked as the fruit of the AI model. Here’s Anthropic’s way of parsing this potentially confusing watermarking approach:

“A detected mark provides a signal that content was processed by Claude, but is not fully conclusive. Detecting a Claude mark tells you that the content may have been processed by Claude. It does not, on its own, confirm the full provenance of the content. For example: Claude may not be the original author. People often use Claude to proofread, translate, summarize, or convert files. The output can carry a Claude mark even if the underlying ideas, text, or data originated from another source.”

Why This Is So Dangerous: I have generally had a positive view of Anthropic’s approach toward human-AI policies. So I’ll admit that I tend to assume they’re usually trying to do the right thing (as much as is possible in a space foundationally based on the unpaid use of prior copyrighted human work). But Anthropic’s watermarking approach has me wondering if this might be, in some way, passive-aggressively deliberate.

The idea of signal dilution through saturation is common in the military, particularly when sending drones out into the battle theater along with chaff (millions of tiny aluminum-coated fibers) to confuse the enemy’s radar tracking systems. Anthropic’s scattershot watermarking method reminds me of this military crypsis-through-commonality strategy. The fact that detecting what is and isn’t AI has become the obsession of many, particularly in fields where writing is a central tool (the courts, journalism, Hollywood), isn’t helping AI companies that are trying to normalize using AI-generated text.

So what can these AI companies do? Well, one solution would be to figure out a way to, through a kind of cultural semiotic collapse, nudge the culture toward no longer even caring whether something was generated by AI or not. Consumer behavior can make or break a company’s bottom line. If human-written text edited by Claude is confoundingly and repeatedly mistaken for Anthropic’s AI-generated text outputs, eventually, the public might stop caring, and as a result, Claude use may inexorably increase. That’s not an accidental policy drift. That’s just good business. (Yes, those last two sentences are me deliberately mimicking AI-style outputs.)

Of course, the only real solution to this quandary is to never put your work into Claude. And if you do, then you certainly shouldn’t subsequently copy that edited work into external documents. But that kind of discipline, in a world where a 1-minute-30-second social media clip is often considered a long, boring documentary, may be asking too much of many users. ✍︎