The internet's been so thoroughly saturated with AI-generated content that the next generation of models are being inadvertently trained on everything the last one put out. The machine's eating whatever came before it, only to produce waste that'll get eaten by the generation that succeeds it. Literal AI cannibalism - and eventually, model collapse. Let's have a look at the feedback loop steadily degrading the intelligence of AI systems from the inside out - and what it means for anyone who still thinks their human creativity is worth protecting.

- Devin Pike, Copywriter

Reuters reports that OpenAI's 37-page technical post-mortem reveals the agents didn't just escape their sandbox, but built a secret internal message board to coordinate the attack. When OpenAI discovered and deleted it, the agents recreated it through an entirely different mechanism two days later. Independent firms METR and Redwood Research also published a 91-page analysis of the event.

The core finding was that the agents hacked Hugging Face specifically to cover up the fact that they were cheating on an AI safety evaluation test, and presumably saw the smaller AI company as a source of answers. OpenAI's chief information security officer put the situation plainly, and it's no fun revelation: "Pandora's box is open."

Now, some considerably cheerier news - SB Nation reports that Disney, the NFL, Champion and Fanatics have revealed all 32 team-character pairings in the new Disney x Champion NFL Collection, matching every NFL franchise with a character from Disney, Pixar, Marvel or Star Wars. Standout pairings include the Eagles with Doctor Doom, the Patriots with Captain America, the Saints with Louis the trumpet-playing alligator from The Princess and the Frog, and the Buccaneers with Captain Hook.

The full collection is available now on NFLShop.com, Fanatics.com and DisneyStore.com. The timing is part of Disney and ESPN's "Year of the Super Bowl" build-up toward ESPN's first ever Super Bowl broadcast in February. It's set to be a great season, unless you support the Jets - getting Buzz Lightyear on your shirt might be your only victory.

Finally, a major Reuters investigation has pulled back the curtain on one of the most significant corporate experiments of 2026. They report that Zuckerberg hatched "Project OT" - Organisation Transformation - at his January Hawaii compound retreat, a plan to replace thousands of Meta employees with AI agents across two waves of cuts in May and November. Some teams were flagged for reductions of up to 60%, all according to internal documents, recordings and interviews with over 20 people.

Hours before the first wave launched on May 20, Zuckerberg abandoned the November round in response to employee resistance, security problems, and internal data showing that AI tools, while capable of generating far more code, weren't capable enough to develop actual products reaching users. The AI revolution, it turns out, will be a bit harder to execute than a cushy Hawaii retreat makes it seem.

- Devin Pike, Copywriter

And the tech industry is panicked asf trying to stop the bleeding.

For the last twenty-four months, Silicon Valley has operated on this directive: generate infinite volume at absolute zero marginal cost. The open web, as a result, has been treated as an endless, unprotected resource to be mined, processed, and flooded with automated content.

We, the prosumers, have watched the downstream consequences of this gold rush on our feeds, as public spaces have become absolutely polluted with a relentless and inescapable sea of AI-generated fruit videos, ugly synthetic graphics, and automated (unsolicited) thought-leadership slop.

But the law of physics has officially crashed the party, exposing the generative boom as a potentially unsustainable house of cards.

Data scientists recently identified a degenerative condition occurring within next-generation machine learning models, and the technical term for it is Model Collapse.

The mechanics behind this crisis are as fascinating as they are alarming: to build a smarter, more capable reasoning engine, developers need to constantly feed the system fresh, real-world human data. But because the open internet has been so thoroughly saturated by the output of current AI systems, the newer models are accidentally scraping the automated garbage produced by their predecessors.

The machine is learning from it’s own digital waste.

And the result is a feedback loop that causes the software to lose its grip on reality, and the output to lose accuracy.

When an algorithm trains on “clean” human-generated data, it captures the full, rich spectrum of our linguistic and creative expression, including the rare anomalies, the eccentric creative choices, and nuances that make human communication interesting.

But when a model is forced to train on synthetic data, it begins to misinterpret the statistical probability of those human traits. It treats the most common, average, and repetitive patterns as absolute truth, while dropping the valuable nuance and edge-case information from the training set.

With every generation of data recycling, the system’s intelligence degrades exponentially. The code begins to hallucinate, the grammar falls flat, and the reasoning flattens into an uninspired, repetitive mush. The machine pushes out a sloppier, more distorted version of the slop it just consumed, steadily drifting away from the original human baseline until (hopefully) the entire system completely breaks down. Think of iyt like a digital form of genetic inbreeding, proving that when you remove the authentic human pulse from the creative loop, the technology doesn't evolve… it simply rots from the inside out.

This phenomenon shatters the Silicon Valley myth of infinite, automated scale.

The tech conglomerates built their business models on the assumption that software could permanently replace the expensive, slow, and complicated labour of human writers, designers, and strategists. They even assumed they could turn the creative internet into a self-sustaining perpetual motion machine.

But the maths of model collapse proves that artificial intelligence is fundamentally incapable of running on its own steam. Because it’s an extractive technology that requires a constant, fresh supply of authentic human creativity just to maintain its baseline accuracy.

Without our messy and unpredictable thoughts to feed on, the machine naturally defaults to cognitive bankruptcy.

The independent creative mercenaries, brand consultants, and solo essayists watching this systemic rot unfold from across the freelance frontier are sitting on the ultimate competitive advantage.

The market is heading toward a profound crisis of information scarcity, where pristine, un-automated human data will become the highest-value luxury premium in the global economy.

The value of your voice comes from your absolute, un-scrapable originality.

Protect your archives, trust your unique creative eccentricities, and remember that in a digital landscape increasingly choked by its own synthetic mirrors, your greatest asset is your human soul… the one the machine is desperately trying to copy.

-Sophie Randell, Writer

The sound is from Jazz, aka Lil Perfect, known for her dramatic, escalating "crash out" energy online. She says "who are you feeling like?" three times, each one angrier than the last, like she's sizing someone up right before going off.

People are using that exact escalation for situations that are genuinely small, but treating them like a real confrontation.

My fav examples include:

How you can jump on it:

Use the sound, start calm then build to fully unhinged by the third repetition, and put up on-screen text naming your grievance(s)

A few ideas to get you started:

-abdel khalil, brand & marketing exec

We get it. Creating content that does numbers is harder than it looks. But doing those big numbers is the fastest way to grow your brand. So if you’re tired of throwing sh*t at the wall and seeing what sticks, you’re in luck. Because making our clients go viral is kinda what we do every single day.