404 Media

2025-12-12 17:06:05

Behind the Blog: Is This Headline 'Clickbait'?

This is Behind the Blog, where we share our behind-the-scenes thoughts about how a few of our top stories of the week came together. This week, we discuss conversational AI, a behind the scenes of the zine, and more.

EMANUEL: I made the terrible mistake of looking at some Hacker News comments this week for my story about a developer whose Google accounts were banned after he uploaded training data to Google Drive. Unbeknownst to him, the training data contained CSAM.

As we’ve explained in previous stories, CSAM is a subject we dread covering not only because it’s one of the most awful things one could think about, but because it’s extremely difficult and legally risky. For understandable reasons, the laws around viewing, let alone possessing CSAM, are strict and punishing, which makes verification for reporting reasons challenging. For similar reasons, it’s something we need to write about very carefully, making sure we don’t wrongfully associate or whitewash someone when it comes to such horrible behavior.

This post is for subscribers only

Become a member to get access to all content
Subscribe now

Behind the Blog: Is This Headline 'Clickbait'?

^{Samantha Cole (404 Media)}

404 Media

2025-12-11 18:10:47

‘Architects of AI’ Wins Time Person of the Year, Sends Gambling Markets Into a Meltdown

The degenerate gamblers of Polymarket and Kalshi who bet that “AI” would win the Time Person of the Year are upset because the magazine has named the “Architects of AI” the person of the year. The people who make AI tools and AI infrastructure are, notably, not “AI” themselves, and thus both Kalshi and Polymarket have decided that people who bet “AI” do not win the bet. On Polymarket alone, people spent more than $6 million betting on AI gracing the cover of Time.

As writer Parker Molloy pointed out, people who bet on AI are pissed. “ITS THE ARCHITECTS OF AI THISNIS [sic] LITERALLY THE BET FUCK KALSHI,” one Kalshi better said.

“This pretty clearly should’ve resolved to yes. If you bought AI, reach out to Kalshi support because ‘AI’ is literally on the cover and in the title ‘Architects of AI.’ They’re not going to change anything unless they hear from people,” said another.

“ThE aRcHiTeCtS oF AI fuck you pay me,” said a third.

“Another misleading bet by Kalshi,” said another gambler. “Polymarket had fair rules and Kalshi did not. They need to fix this.”

But bag holders on Polymarket are also pissed. “This is a scam. It should be resolved to a cancellation and a full refund to everyone,” said a gambler who’d put money down on Jensen Huang and lost. Notably, on Kalshi, anyone who bet on any of the “Architects of AI,” won the bet (meaning Sam Altman, Elon Musk, Jensen Huang, Dario Amodei, Mark Zuckerberg, Lisa Su, and Demis Hassabis), while anyone who bet their products—“ChatGPT” and “OpenAI” did not win. On Polymarket, the rules were even more strict, i.e. people who bet “Jensen Huang” lost but people who bet “Other” won.

“FUCK YOU FUCKING FUCK Shayne Coplan [CEO of Polymarket],” said someone who lost about $50 betting on AI to make the cover.

Polymarket made its reasoning clear in a note of “additional context” on the market.

“This market is about the person/thing named as TIME's Person of the Year for 2025, not what is depicted on the cover. Per the rules, “If the Person of the Year is ‘Donald Trump and the MAGA movement,’ this would qualify to resolve this market to ‘Trump.’ However if the Person of the Year is ‘The MAGA movement,’ this would not qualify to resolve this market to ‘Trump’ regardless of whether Trump is depicted on the cover,” it said.

“Accordingly, a Time cover which lists ‘Architects of AI’ as the person of the year will not qualify for ‘AI’ even if the letters ‘AI’ are depicted on the cover, as AI itself is not specifically named.”

It should be noted how incredibly stupid all of this is, which is perhaps appropriate for the year 2025, in which most of the economy consists of reckless gambling on AI. People spent more than $55 million betting on the Time Person of the Year on Polymarket, and more than $19 million betting on the Time Person of the Year on Kalshi. It also presents one of the many downsides of spending money to bet on random things that happen in the world. One of the most common and dumbest things that people continue to do to this day despite much urging otherwise is anthropomorphize AI, which is distinctly not a person and is not sentient.

Time almost always actually picks a “person” for its Person of the Year cover, but it does sometimes get conceptual with it, at times selecting groups of people (“The Silence Breakers” of the #MeToo movement, the “Whistleblowers,” the “Good Samaritans,” “You,” and the “Ebola Fighters,” for example). In 1982 it selected “The Computer” as its “Machine of the Year,” and in 1988 it selected “The Endangered Earth” as “Planet of the Year.”

Polymarket’s users have been upset several times over the resolution of bets in the past few weeks and their concerns highlight how easy it is to manipulate the system. In November, an unauthorized edit of a live map of the Ukraine War allowed gamblers to cash in on a battle that hadn’t happened. Earlier this month, a trader made $1 million in 24 hours betting on the results of Google’s 2025 Year In Search Rankings and other users accused him of having inside knowledge of the process. Over the summer, Polymarket fought a war over whether or not President Zelenskyy had worn a suit. Surely all of this will continue to go well and be totally normal moving forward, especially as these prediction markets begin to integrate themselves with places such as CNN.

Kalshi to become CNN’s official prediction market partner

Kalshi’s data will serve as a powerful complement to CNN’s reporting.

^Kalshi

404 Media

2025-12-11 16:48:39

Disney Invests $1 Billion in the AI Slopification of Its Brand

The first thing I saw this morning when I opened X was an AI-generated trailer for Avengers: Doomsday. Robert Downey Jr’s Doctor Doom stood in a shapeless void alongside Captain America and Reed Richards. It was obvious slop but it was also close in tone and feel of the last five years of Disney’s Marvel movies. As media empires consolidate, nostalgia intensifies, and AI tools spread, Disney’s blockbusters feel more like an excuse to slam recognizable characters together in a contextless morass.

So of course Disney has announced it signed a deal with OpenAI today that will soon allow fans to make their own officially licensed Disney slop using Sora 2. The house that mouse built, and which has been notoriously protective of its intellectual property, opened up the video generator, saw the videos featuring Nazi Spongebob and criminal Pikachu, and decided: We want in.

According to a press release, the deal is a 3 year licensing agreement that will allow the AI company’s short form video platform Sora to generate slop videos using characters like Mickey Mouse and Iron Man. As part of the agreement, Disney is investing $1 billion of equity into OpenAI, said it will become a major customer of the company, and promised that fan and corporate AI-generated content would soon come to Disney+, meaning that Disney will officially begin putting AI slop into its flagship streaming product.

The deal extends to ChatGPT as well and, starting in early 2026, users will be able to crank out officially approved Disney slop on multiple platforms. When Sora 2 launched in October, it had little to no content moderation or copyright guidelines and videos of famous franchise characters doing horrible things flooded the platform. Pikachu stole diapers from a CVS, Rick and Morty pushed crypto currencies, and Disney characters shouted slurs in the aisles of Wal-Mart.

It is worth mentioning that, although Disney has traditionally been extremely protective of its intellectual property, the company’s princesses have become one of the most common fictional subjects of AI porn on the internet; 404 Media has found at least three different large subreddits dedicated to making AI porn of characters like Elsa, Snow White, Rapunzel, and Tinkerbell. In this case, Disney is fundamentally throwing its clout behind a technology that has thus far most commonly been used to make porn of its iconic characters.

After the hype of the launch, OpenAI added an “opt-in” policy to Sora that was meant to prevent users from violating the rights of copyright holders. It’s trivial to break this policy however, and circumvent the guardrails preventing a user from making a lewd Mickey Mouse cartoon or episode of The Simpsons. The original sin of Sora and other AI systems is that the training data is full of copyrighted material and the models cannot be retrained without great cost, if at all.

If you can’t beat the slop, become the slop.

“The rapid advancement of artificial intelligence marks an important moment for our industry, and through this collaboration with OpenAI we will thoughtfully and responsibly extend the reach of our storytelling through generative AI, while respecting and protecting creators and their works,” Bob Iger, CEO of Disney, said in the press release about the agreement.

The press release explained that Sora users will soon have “official” access to 200 characters in the Disney stable, including Loki, Thanos, Darth Vader, and Minnie Mouse. In exchange, Disney will begin to use OpenAI’s APIs to “build new products” and it will deploy “ChatGPT for its employees.”

I’m imagining a future where AI-generated fan trailers of famous characters standing next to each other in banal liminal spaces is the norm. People have used Sora 2 to generate some truly horrifying videos, but the guardrails have become more aggressive. As Disney enters the picture, I imagine the platform will become even more anodyne. Persistent people will slip through and generate videos of Goofy and Iron Man sucking and fucking, sure, but the vast majority of what’s coming will be safe corporate gruel that resembles a Marvel movie.

OpenAI’s Sora 2 Copyright Infringement Machine Features Nazi SpongeBobs and Criminal Pikachus

The main use of Sora appears to generate brainrot of major beloved copyrighted characters, to say nothing of the millions of articles, images, and videos OpenAI has scraped.

^{Jason Koebler (404 Media)}

404 Media

2025-12-11 15:56:21

Porn Is Being Injected Into Government Websites Via Malicious PDFs

Dozens of government and university websites belonging to cities, towns, and public agencies across the country are hosting PDFs promoting AI porn apps, porn sites, and cryptocurrency scams; dozens more have been hit with a website redirection attacks which lead to animal vagina sex toy ecommerce pages, penis enlargement treatments, automatically-downloading Windows program files, and porn.

“Sex xxx video sexy Xvideo bf porn XXX xnxx Sex XXX porn XXX blue film Sex Video xxx sex videos Porn Hub XVideos XXX sexy bf videos blue film Videos Oficial on Instagram New Viral Video The latest original video has taken the internet by storm and left viewers in on various social media platforms ex Videos Hot Sex Video Hot Porn viral video,” reads the beginning of a three-page PDF uploaded to the website of the Irvington, New Jersey city government’s website.

The PDF, called “XnXX Video teachers fucking students Video porn Videos free XXX Hamster XnXX com” is unlike many of the other PDFs hosted on the city’s website, which include things like “2025-10-14 Council Minutes,” “Proposed Agenda 9-22-25,” and “Landlord Registration Form (1 & 2 unit dwelling).”

It is similar, however, to another PDF called “30 Best question here’s,” which looks like this:

Irvington, which is just west of Newark and has a population of 61,000 people, has fallen victim to an SEO spam attack that has afflicted local and state governments and universities around the United States.

💡
Do you know anything else about whatever is going on here? I would love to hear from you. Using a non-work device, you can message me securely on Signal at jason.404. Otherwise, send me an email at jason@404media.co.

Researcher Brian Penny has identified dozens of government and university websites that hosted PDF guides for how to make AI porn, PDFs linking to porn videos, bizarre crypto spam, sex toys, and more.

Reginfo.gov, a regulatory affairs compliance website under the federal government’s General Services Administration, is currently hosting a 12 page PDF called “Nudify AI Free, No Sign-Up Needed!,” which is an ad and link to an abusive AI app designed to remove a person’s clothes. The Kansas Attorney General’s office and the Mojave Desert Air Quality Management District Office in California hosted PDFs called “DeepNude AI Best Deepnude AI APP 2025.” Penny found similar PDFs on the websites for the Washington Department of Fish and Wildlife, the Washington Fire Commissioners Association, the Florida Department of Agriculture, the cities of Jackson, Mississippi and Massillon, Ohio, various universities throughout the country, and dozens of others. Penny has caught the attention of local news throughout the United States, who have reported on the problem.

The issue appears to be stemming from websites that allow people to upload their own PDFs, which then sit on these government websites. Because they are loaded with keywords for widely searched terms and exist on government and university sites with high search authority, Google and other search engines begin to surface them. In the last week or so, many (but not all) of the PDFs Penny has discovered have been deleted by local governments and universities.

But cities seem like they are having more trouble cleaning up another attack, which is redirecting traffic from government URLs to porn, e-commerce, and spam sites. In an attack that seems similar to what we reported in June, various government websites are somehow being used to maliciously send traffic elsewhere. For example, the New York State Museum’s online exhibit for something called “The Family Room” now has at least 11 links to different types of “realistic” animal vagina pocket masturbators, which include “Zebra Animal Vagina Pussy Male Masturbation Cup — Pocket Realistic Silicone Penis Sex Toy ($27.99),” and “Must-have Horse Pussy Torso Buttocks Male Masturbator — Fantasy Realistic Animal Pussie Sex Doll.”

Links Penny found on Knoxville, Tennessee’s site for permitting inspections first go to a page that looks like a government site for hosting files then redirects to a page selling penis growth supplements that features erect penises (human penises, mercifully), blowjobs, men masturbating, and Dr. Oz’s face.

Another Knoxville link I found, which purports to be a pirated version of the 2002 Vin Diesel film XXX simply downloaded a .exe file to my computer.

Penny believes that what he has found is basically the tip of the iceberg, because he is largely finding these by typing things like “nudify site:.gov” “xxx site:.gov” into Google and clicking around. Sometimes, malicious pages surface only on image searches or video searches: “Basically the craziest things you can think of will show up as long as you’re on image search,” Penny told 404 Media. “I’ll be doing this all week.”

The Nevada Department of Transportation told 404 Media that “This incident was not related to NDOT infrastructure or information systems, and the material was not hosted on NDOT servers.This unfortunate incident was a result of malicious use of a legitimate form created using the third-party platform on which NDOT’s website is hosted. NDOT expeditiously worked with our web hosting vendor to ensure the inappropriate content was removed.” It added that the third-party is Granicus, a massive government services company that provides website backend infrastructure for many cities and states around the country, as well as helps them stream and archive city council meetings, among other services. Several of the affected local governments use Granicus, but not all of them do; Granicus did not respond to two requests for comment from 404 Media.

The California Secretary of State’s Office told 404 Media: “A bad actor uploaded non-business documents to the bizfile Online system (a portal for business filings and information). The files were then used in external links allowing public access to only those uploaded files. No data was compromised. SOS staff took immediate action to remove the ability to use the system for non-SOS business purposes and are removing the unauthorized files from the system.” The Washington Department of Fish and Wildlife said “WDFW is aware of this issue and is actively working with our partners at WaTech to address it.” The other government agencies mentioned in this article did not respond to our requests for comment.

AI porn on state websites

Mainland whistleblower warns of malicious webpages being hosted on government websites nationwide, including in Hawaiʻi.

^{Michael Brestovansky (Aloha State Daily)}

404 Media

2025-12-11 14:17:04

Scientists Discover the Earliest Human-Made Fire, Rewriting Evolutionary History

🌘
Subscribe to 404 Media to get The Abstract, our newsletter about the most exciting and mind-boggling science news and studies of the week.

Humans made fires as early as 400,000 years ago, pushing the timeline of this crucial human innovation back a staggering 350,000 years, reports a study published on Wednesday in Nature.

Mastery of fire is one of the most significant milestones in our evolutionary history, enabling early humans to cook nutritious food, seek protection from predators, and establish comfortable spaces for social gatherings. The ability to make fires is completely unique to the Homo genus that includes modern humans (Homo sapiens) and extinct humans, including Neanderthals.

Early humans may have opportunistically exploited wildfires more than one million years ago, but the oldest known controlled fires, which were intentionally lit with specialized tools, were previously dated back to about 50,000 years ago at Neanderthal sites in France.

Now, archaeologists have unearthed the remains of campfires ignited by an unidentified group of humans 400,000 years ago at Barnham, a village near the southern coast of the United Kingdom.

“This is a 400,000-year-old site where we have the earliest evidence of making fire—not just in Britain or Europe, but in fact, anywhere else in the world,” said Nick Ashton, an archaeologist at the British Museum who co-authored the study, in a press briefing held on Tuesday.

“Many of the great turning points in human development, and the development of our civilization, depended on fire,” added co-author Rob Davis, also an archaeologist at the British Museum. “We're a species who have used fire to really shape the world around us—in belief systems, as well. It's a very prominent part of belief systems across the world.”

Artifacts have been recovered from Barnham for more than a century, but the remnants of this ancient hearth were identified within the past decade. The researchers were initially tipped off by the remains of heated clay sediments, hydrocarbons associated with fire, and fire-cracked flint handaxes.

But the real smoking gun was the discovery of two small fragments of iron pyrite, a mineral commonly used to strike flint to produce sparks at later prehistoric campfires such as the French Neanderthal sites.
Discovery of the first fragment of iron pyrite in 2017 at Barnham, Suffolk Image: Jordan Mansfield, Pathways to Ancient Britain Project.
“Iron pyrite is a naturally occurring mineral, but through geological work in the area over the last 36 years, looking at 26 sites, we argue that pyrite is incredibly rare in the area,” said Ashton. “We think humans brought pyrite to the site with the intention of making fire.”

The fire-starters were probably Neanderthals, who were known to be present in the region at the time thanks to a skull found in Swanscombe, about 80 miles northeast of Barnham. But it’s possible that the fires were made by another human lineage such as Homo heidelbergensis, which also left bones in the U.K. around the same period. It was not Homo sapiens as our lineage emerged in Africa later, about 300,000 years ago.

Regardless of this group’s identity, its ability to make fire would have been a major advantage, especially in the relatively cold environment of southern Britain at the time. It also hints that the ability to make fire extends far deeper into the past than previously known.

“We assume that the people who made the fire at Barnham brought the knowledge with them from continental Europe,” said co-author Chris Stringer, a physical anthropologist at the Natural History Museum. “There was a land bridge there. There had been a major cold stage about 450,000 years ago, which had probably wiped out everyone in Britain. Britain had to be repopulated all over again.”

“Having that use of fire, which they must have brought with them when they came into Britain, would have helped them colonize this new area and move a bit further north to places where the winters are going to be colder,” he continued. “You can keep warm. You can keep wild animals away. You get more nutrition from your food.”
Excavation of the ancient campfire, removing diagonally opposed quadrants. The reddened sediment between band B’ is heated clay. Image: Jordan Mansfield, Pathways to Ancient Britain Project.
Although these humans likely had brains close in size to our own, the innovation of controlled fire would have amplified their cognitive development, social bonds, and symbolic capacities. In the flickering light of ancient campfires, these humans shared food, protection, and company, passing on a tradition that fundamentally reshaped our evolutionary trajectory.

“People were sitting around the fires, sharing information, having extra time beyond pure daylight to make things, to teach things, to communicate with each other, to tell stories,” Stringer said. “Maybe it may have even fueled the development of language.”

“We've got this crucial aspect in human evolution, and we can put a marker down that it was there 400,000 years ago,” he concluded.

Earliest evidence of making fire - Nature

Baked sediment, heat-shattered artefacts and introduced pyrite in a 400,000-year-old Palaeolithic occupation site in Suffolk, UK provide evidence of intentional fire-making, marking a pivotal moment in human development.

^Nature