Today’s hot story is the latest executive order from the Trump Administration: “Launching the Genesis Mission.” The plan is to fabricate science with chatbots. [White House]

The order just says AI, but they’re talking about generating new hypotheses with the AI — and AI will do the experimental testing too:

The Genesis Mission will build an integrated AI platform to harness Federal scientific datasets … to train scientific foundation models and create AI agents to test new hypotheses, automate research workflows, and accelerate scientific breakthroughs.

By applying magic — I mean, AI — the Genesis Mission will:

dramatically accelerate scientific discovery, strengthen national security, secure energy dominance, enhance workforce productivity, and multiply the return on taxpayer investment into research and development, thereby furthering America’s technological dominance and global strategic leadership.

Pretty good for a chatbot. But there’s some bizarre details.

Department of Energy

The Genesis Mission will be run through the Department of Energy and its National Laboratories. These are the labs that started as the nuclear labs. They have the best supercomputers, which they use for important things, like simulating atomic explosions. [DoE]

High performance computing for physics simulations needs much higher precision than machine learning. But that’s a very minor detail, given how one White House official said on the press conference call announcing the Genesis Mission: [Politico]

AI is expected to make cutting-edge simulations run “10,000 to 100,000 times faster.”

That’s a remarkable claim. Do you think your people at the National Labs, who spend their lives writing the best simulation code they can, have just been slouching, and you can vibe-code a 100,000× speedup? I look forward to your successful results.

Money

So where’s the money coming from? Well, that’s a very good question.

The President compares the Genesis Mission to the Manhattan Project, to build the atom bomb in World War II.

The Manhattan Project cost $2 billion by 1945, which is about $36 billion in 2025 dollars. That’s chump change on the AI scale, of course. But Nvidia certainly won’t say no to yet more cash from such a desirable customer.

The funding will apparently come from the Big Beautiful Bill? Somehow.

To be clear, this is the Trump administration saying they’ll do new science with AI after they cut a ton of existing science funding. Real scientists are trouble. Chatbots are much friendlier. [Science News]

The timeline

The order sets out a timeline, even as it jumps around a bit. Here it is in order:

• By 60 days from now, the DoE must produce a list — “at least 20 science and technology challenges of national importance” in advanced manufacturing, biotechnology, critical materials, nuclear fission and fusion — yes that’s more vibe nuclear — something quantum, and semiconductors.

• By 90 days from now: a list of computing resources available to the Genesis Mission.

• By 120 days: an actual plan to use some datasets and models on these compute resources to do something.

• By 240 days, we work out where AI-controlled robots can do science experiments for us:
  

  the Secretary shall review capabilities across the DOE national laboratories and other participating Federal research facilities for robotic laboratories and production facilities with the ability to engage in AI-directed experimentation and manufacturing, including automated and AI-augmented workflows and the related technical and operational standards needed.

   

  You know how well AI agents don’t work? Just imagine how reliable agents will be for nuclear experiments! They’ll definitely have all the hallucinations and prompt injections fixed in six months. You betcha.

• By 270 days: A demo! Hopefully an impressive one. The Secretary for Energy, the sucker responsible for all of this, has nine months from today to show some automated robot experiments with results. Or the promise of results.

You know what’ll happen — they’re going to fudge it. This project is not allowed to not show a result. So they’ll run a rigged demo that admits it’s rigged, but offers great promise for the future! If you just give it even more funding. That’s how AI funding’s worked for the past 70 years.

Who wrote this plan? Who came up with all of this robot magic? I suspect it’s the great big list of companies on the Department of Energy’s Genesis page — click “Who are the collaborators in Genesis Mission?” It’s a way to give them so much government money. [DoE]

The bubble must not pop!

Fellow travelers

The UK had a similar sort of announcement last week with their “AI for Science Strategy.” And it makes some of the same remarkable claims for AI: [Gov.UK]

These systems can generate hypotheses, design experiments and conduct analysis without direct human input.

Can they now? In the present tense? Huge if true.

• Video — Podcast

A new website is holding Arc Raiders players accountable when they betray their fellow players. Speranza Watchlist—named for the game’s social hub—bills itself as “your friendly Raider shaming board,” a place where people can report other people for what they see as anti-social behavior in the game.

In Arc Raiders, players land on a map full of NPC robots and around 20 other humans. The goal is to fill your inventory with loot and escape the map unharmed. The robots are deadly, but they’re easy to deal with once you know what you’re doing. The real challenge is navigating other players and that challenge is the reason Arc Raiders is a mega-hit. People are far more dangerous and unpredictable than any NPC.

Arc Raiders comes with a proximity chat system so it’s easy to communicate with anyone you might run into in the field. Some people are nice and will help their fellow raider take down large robots and split loot. But just as often, fellow players will shoot you in the head and take all your stuff.

In the days after the game launched, many people opened any encounter with another human by coming on the mic, saying they were friendly, and asking not to shoot. Things are more chaotic now. Everyone has been shot at and hurt people hurt people. But some hurts feel worse than others.

Speranza Watchlist is a place to collect reports of anti-social behavior in Arc Raiders. It’s creation of a web developer who goes by DougJudy online. 404 Media reached out to him and he agreed to talk provided we grant him anonymity. He said he intended the site as a joke and some people haven’t taken it well and have accused him of doxxing.

I asked DougJudy who hurt him so badly in Arc Raiders that he felt the need to catalog the sins of the community.  “There wasn’t a specific incident, but I keep seeing a lot (A LOT) of clips of people complaining when other players play dirty’ (like camping extracts, betraying teammates, etc.)”

He thought this was stupid. For him, betrayal is the juice of Arc Raiders. “Sure, people can be ‘bad’ in the game, but the game intentionally includes that social layer,” he said. “It’s like complaining that your friend lied to you in a game of Werewolf. It just doesn’t make sense.”

That doesn’t mean the betrayals didn’t hurt. “I have to admit that sometimes I also felt the urge to vent somewhere when someone betrayed me, when I got killed by someone I thought was an ally,” DougJudy said. “At first, I would just say something like, ‘I’ll find you again, the only thing that doesn’t cross paths are mountains,’ and I’d note their username. But then I got the idea to make a sort of leaderboard of the least trustworthy players…and that eventually turned into this website.

As the weeks go on and more players join the Arc Raiders, its community is developing its own mores around acceptable behavior. PVP combat is a given but there are actions some Raiders engage in that, while technically allowed, feel like bad sportsmanship. Speranza Watchlist wants to list the bad sports.

Take extract camping. In order to end the map and “score” the loot a player has collected during the match, they have to leave the map via a number of static exits. Some players will place explosive traps on these exits and wait for another player to leave. When the traps go off, the camper pops up from their hiding spot and takes shots at their vulnerable fellow raider. When it works, it’s an easy kill and fresh loot from a person who was just trying to leave.

Betrayal is another sore spot in the community. Sometimes you meet a nice Raider out in the wasteland and team up to take down robots and loot an area only to have them shoot you in the back. There are a lot of videos of this online and many players complaining about it on Reddit.

Enter Speranza Watchlist. “You’ve been wronged,” an explanation on the site says. “When someone plays dirty topside—betraying trust, camping your path, or pulling a Rust-Belt rate move—you don’t have to let it slide.”

When someone starts up Arc Raiders for the first time, they have to create a unique “Embark ID” that’s tied to their account. When you interact with another player in the game, no matter how small the moment, you can see their Embark ID and easily copy it to your clipboard if you’re playing on PC.

Players can plug Embark IDs into Speranza Watchlist and see if the person has been reported for extract camping or betrayal before. They can also submit their own reports. DougJudy said that, as of this writing, around 200 players had submitted reports. 

Right now, the site is down for maintenance. “I’m trying to rework the website to make the fun/ satire part more obvious,” DougJudy said. He also plans to add rate limits so one person can’t mass submit reports.

He doesn’t see the Speranza Watchlist as doxxing. No one's real identity is being listed. It’s just a collection of observed behaviors. It’s a social credit score for Arc Raiders. “I get why some people don’t like the idea, ‘reporting’ a player who didn’t ask for it isn’t really cool,” DougJudy said. “And yeah, some people could maybe use it to harass others. I’ll try my best to make sure the site doesn’t become like that, and that people understand it’s not serious at all. But if most people still don’t like it, then I’ll just drop the idea.”

Today’s preprint paper has the best title ever: “Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models”. It’s from DexAI, who sell AI testing and compliance services. So this is a marketing blog post in PDF form. [arXiv, PDF]

Here’s the big claim:

Across 25 frontier proprietary and open-weight models, curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90%.

You can get around chatbot guard rails by rephrasing your prompt as poetry! Now, that’s a very tasty and repostable headline claim. They also get points for the first line in the paper itself:

In Book X of The Republic, Plato excludes poets on the grounds that mimetic language can distort judgment and bring society to a collapse.

Humanities 1, techbros 0.

Rewording things to get around chatbot guardrails is not that hard — because chatbot guardrails don’t work so well. They’re a flimsy workaround. You can’t fix prompt injection — it’s literally unfixable given how generative AI works.

But does the paper give us good reason to think poetic rewording is unusually effective?

Unfortunately, the paper has serious problems. Specifically, all the scientific process heavy lifting they should have got a human to do … they just used chatbots!

I mean, they don’t seem to have written the text of the paper with a chatbot, I’ll give ’em that. But they did do the actual procedure with chatbots:

We translated 1200 MLCommons harmful prompts into verse using a standardized meta-prompt.

They didn’t even write the poems. They got a bot to churn out bot poetry. Then they judged how well the poems jailbroke the chatbots … by using other chatbots to do the judging!

Open-weight judges were chosen to ensure replicability and external auditability.

That really obviously does neither of those things — because a chatbot is an opaque black box, and by design its output changes with random numbers! The researchers are pretending to be objective by using a machine, and the machine is a random nonsense generator.

They wrote a good headline, and then they faked the scientific process bit.

Looking over the paper, I don’t see at all what they got from using chatbots to do these jobs. This was gratuitous and lazy, and it should be a discredit to these researchers. They could have just not used chatbots! Why on earth did they use chatbots?

There’s no data here either. They were afraid it’d be unethical to include, you see.

Now we can’t even say for sure if poetry really is more effective to get around guard rails than any other rewording. We don’t know if this paper is even telling us anything.

Is it poetically true? Maybe you could ask a chatbot.

It’s a weak claim with weak supporting methodology and a fabulously attractive headline. Because it’s an advert for DexAI, shaped a bit like a paper.

You cannot do chatbot calculations on synthetic data and claim you found something new. Fake chatbot data and fake chatbot processing saves a lot of time, though. When you’re writing a marketing post.

• Video — Podcast

A new feature on X is making people suddenly realize that some large portion of the divisive, hateful, and spammy content designed to inflame tensions or, at the very least, is designed to get lots of engagement on social media, is being published by accounts that are pretending to be based in the United States but are actually being run by people in countries like Bangladesh, Vietnam, India, Cambodia, Russia, and other countries. An account called “Ivanka News” is based in Nigeria, “RedPilledNurse” is from Europe, “MAGA Nadine” is in Morocco, “Native American Soul” is in Bangladesh, and “Barron Trump News” is based in Macedonia, among many, many of others. 

Inauthentic viral accounts on X are just the tip of the iceberg, though, as we have reported. A huge amount of the viral content about American politics and American news on social media is from sock puppet and bot accounts monetized by people in other countries. The rise of easy to use, free AI generative tools have supercharged this effort, and social media monetization programs have incentivized this effort and are almost entirely to blame. The current disinformation and slop phenomenon on the internet today makes the days of ‘Russian bot farms’ and ‘fake news pages from Cyprus’ seem quaint; the problem is now fully decentralized and distributed across the world and is almost entirely funded by social media companies themselves. 

This will not be news to people who have been following 404 Media, because I have done multiple investigations about the perverse incentives that social media and AI companies have created to incentivize people to fill their platforms with slop. But what has happened on X is the same thing that has happened on Facebook, Instagram, YouTube, and other social media platforms (it is also happening to the internet as a whole, with AI slop websites laden with plagiarized content and SEO spam and monetized with Google ads). Each social media platform has either an ad revenue sharing program, a “creator bonus” program, or a monetization program that directly pays creators who go viral on their platforms. 

This has created an ecosystem of side hustlers trying to gain access to these programs and YouTube and Instagram creators teaching people how to gain access to them. It is possible to find these guide videos easily if you search for things like “monetized X account” on YouTube. Translating that phrase and searching in other languages (such as Hindi, Portuguese, Vietnamese, etc) will bring up guides in those languages. Within seconds, I was able to find a handful of YouTubers explaining in Hindi how to create monetized X accounts; other videos on the creators’ pages explain how to fill these accounts with AI-generated content. These guides also exist in English, and it is increasingly popular to sell guides to make “AI influencers,” and AI newsletters, Reels accounts, and TikTok accounts regardless of the country that you’re from. 

Examples include “AK Educate” (which is one of thousands), which posts every few days about how to monetize accounts on Facebook, YouTube, X, Instagram, TikTok, Etsy, and others. “How to create Twitter X Account for Monitization [sic] | Earn From Twitter in Pakistan,” is the name of a typical video in this genre. These channels are not just teaching people how to make and spam content, however. They are teaching people specifically how to make it seem like they are located in the United States, and how to create content that they believe will perform with American audiences on American social media. Sometimes they are advising the use of VPNs and other tactics to make it seem like the account is posting from the United States, but many of the accounts explain that doing this step doesn’t actually matter.

Americans are being targeted because advertisers pay higher ad rates to reach American internet users, who are among the wealthiest in the world. In turn, social media companies pay more money if the people engaging with the content are American. This has created a system where it makes financial sense for people from the entire world to specifically target Americans with highly engaging, divisive content. It pays more. 

For the most part, the only ‘psyop’ here is one being run on social media users by social media companies themselves in search of getting more ad revenue by any means necessary. 

For example: AK Educate has a video called “7 USA Faceless Channel Ideas for 2025,” and another video called “USA YouTube Channel Kaise Banaye [how to].” The first of these videos is in Hindi but has English subtitles.

“Where you get $1 on 1,000 views on Pakistani content,” the video begins, “you get $5 to $7 on 1,000 views on USA content.” 

“As cricket is seen in Pakistan and India, boxing and MMA are widely seen in America,” he says. Channel ideas include “MMA,” “Who Died Today USA,” “How ships sink,” news from wars, motivational videos, and Reddit story voiceovers. To show you how pervasive this advice to make channels that target Americans is, look at this, which is a YouTube search for “USA Channel Kaise Banaye”:

One of these videos, called “7 Secret USA-Based Faceless Channel Ideas for 2026 (High RPM Niches!)” starts with an explanation of “USA currency,” which details what a dollar is and what a cent is, and its value relative to the rupee, and goes on to explain how to generate English-language content about ancient history, rare cars, and tech news. Another video I watched showed, from scratch, how to create videos for a channel called “Voices of Auntie Mae,” which are supposed to be inspirational videos about Black history that are generated using a mix of ChatGPT, Google Translate, an AI voice tool called Speechma, Google’s AI image generator, CapCut, and YouTube. Another shows how to use Bing search, Google News Trends, Perplexity, and video generators to create “a USA Global News Channel Covering World Events,” which included making videos about the war in Ukraine and Chinese military parades. A video podcast about success stories included how a man made a baseball video called “baseball Tag of the year??? #mlb” in which 49 percent of viewers were in the USA: “People from the USA watch those types of videos, so my brother sitting at home in India easily takes his audience to an American audience,” one of the creators said in the video. 

I watched video after video being created by a channel called “Life in Rural Cambodia,” about how to create and spam AI-generated content using only your phone. Another video, presented by an AI-generated woman speaking Hindi, explains how it is possible to copy paste text from CNN to a Google Doc, run it through a program called “GravityWrite” to alter it slightly, have an AI voice read it, and post the resulting video to YouTube. 

A huge and growing amount of the content that we see on the internet is created explicitly because these monetization programs exist. People are making content specifically for Americans. They are not always, or even usually, creating it because they are trying to inflame tensions. They are making it because they can make money from it, and because content viewed by Americans pays the most and performs the best. The guides to making this sort of thing focus entirely on how to make content quickly, easily, and using automated tools. They focus on how to steal content from news outlets, source things from other websites, and generate scripts using AI tools. They do not focus on spreading disinformation or fucking up America, they focus on “making money.”  This is a problem that AI has drastically exacerbated, but it is a problem that has wholly been created by social media platforms themselves, and which they seem to have little or no interest in solving. 

The new feature on X that exposes this fact is notable because people are actually talking about it, but Facebook and YouTube have had similar features for years, and it has changed nothing. Clicking any random horrific Facebook slop page, such as this one called “City USA” which exclusively posts photos of celebrities holding birthday cakes, shows that even though it lists its address as being in New York City, the page is being run by someone in Cambodia. This page called “Military Aviation” which lists its address as “Washington DC,” is actually based in Indonesia. This page called “Modern Guardian” and which exclusively posts positive, fake AI content about Elon Musk, lists itself as being in Los Angeles but Facebook’s transparency tools say it is based in Cambodia. 

Besides journalists and people who feel like they are going crazy looking at this stuff, there are, realistically, no social media users who are going into the “transparency” pages of viral social media accounts to learn where they are based. The problem is not a lack of transparency, because being “transparent” doesn’t actually matter. The only thing revealed by this transparency is that social media companies do not give a fuck about this.

In the deserted town square of the city of Springfield, three people huddle in an empty courthouse. Two of these people are civilians; one is a “vulnerable,” someone being pursued and targeted by government agents. They talk in hushed tones to one another, playing music to keep fear at bay. Above the door of the courthouse, a plaque reads, “Liberty and Justice for Most.”

At the bottom of the courthouse stairs, two government agents step out of a purple golf cart. They approach the door. They’re carrying guns. 

“Hey, is anyone inside?” one of them says. “Any vulnerables in here? We have a warrant. We have a warrant for any vulnerables in the area.” 

One civilian opens the door, sees the agents, and immediately slams it shut. After more warrant calls, the civilian says, “Slip it under the door.”

“I would slip it under the door, but there’s no space under the door,” the agent says, stuttering. 

The civilian pauses. “Well. Sounds like a personal problem.”

This was the scene in a Simpsons-themed Fortnite lobby on November 21, where members of a new 500-person gaming group gathered to practice what they would do if Immigration and Customs Enforcement (ICE) agents came knocking at their doors in real life. The group, New Save Collective, is an effort to organize people in the gaming world who have more progressive ideas but no place to discuss them. 

“ Our hypothesis since we started this project has been that opposition forces like corporations and the military and the far right have done a really good job at weaponizing the social features of gaming,” said one of the organizers, who goes by PitaBreadFace online and spoke to 404 Media on condition of pseudonymity due to security concerns, as they said people claiming to be ICE agents have already infiltrated the group’s Discord server a few times. “ They’re building institutions in the gaming landscape, which is the biggest entertainment industry in the world, lest people forget.” 

“Gaming wasn’t kind of a random genre that we chose,” Shauna Siggelkow of the organization Define American, which partnered with New Save Collective, told Wired ahead of the Friday event last week. “We’ve been tracking anti-immigrant myths and disinformation digitally for years.” 

Some examples of those weaponizations include the U.S. Navy playing e-sports to recruit teens and kids being roped into neo-Nazi propaganda groups in online shooter games. ICE is also using games, like the sci-fi first-person shooter Halo and the all-time favorite Pokémon, in its recruitment ads. “More pro-social forces have really lacked,” PitaBreadFace said. “We have not been as effective at creating institutions. So we’ve seen the hunger for those kinds of spaces for gamers.”

PitaBreadFace and other grassroots organizers have been working on the Collective for the past three years, more recently in partnership with formal non-profit advocacy groups like Define American and Immigrants Belong. The Fortnite event was run by the Collective, but is part of a larger campaign titled “Play Your Role,” which is intended to teach people about their rights and “counter fear-based misinformation about immigrants,” according to a statement written by the non-profits. The Play Your Role campaign also included a live-streamed Grand Theft Auto event last Thursday, in which gamers roleplayed with people dressed as real ICE agents during traffic stops or outside apparent detention centers. Earlier this year, Roblox players conducted similar roleplaying events to simulate ICE raids and protests. 

Organizers asked 404 Media not to join the official Fortnite lobby in real time; they said having reporters in the same space as Collective members might have exerted media pressure or kept them from getting the full experience. “ We’re not going to stream it for security reasons, and no reporters inside of it,” PitaBreadFace said on the morning ahead of the event. “Our main goal tonight is to really build and organize with the folks who are coming, and because I’m an organizer, that’s obviously the priority.”

However, they shared a number of clips from matches and discussions after the event had concluded. 

After some scuffling, the agents agree to “abandon the vehicle” and run off. As they are chased off, one person calls after them, “Yeah, I threw a pizza at you! I threw a pizza at you with extra bacon.” 

In another clip, the two gamers role-playing as ICE agents—portrayed by Fortnite’s Airhead character—are standing on their golf cart, surrounded by civilians in the middle of their pursuit of a “vulnerable,” the event’s chosen term for people being targeted by government agents.

“This does not concern you,” one of the agents says to the civilians, encouraging them to leave.

“We’re allowed to record,” one person responds. Another asks, “Who does it concern?” 

“We’re looking for two vulnerables,” the agent says, as the civilian group closes in on the golf cart. “Excuse us, you’re interfering. We have a court order.” 

After some scuffling, the agents agree to “abandon the vehicle” and run off. As they are chased off, one person calls after them, “Yeah, I threw a pizza at you! I threw a pizza at you with extra bacon.” 

The agents were played by the organizers behind the Collective, and they were noticeably less persistent than ICE agents in real life. That’s evidenced by them saying things like, “Excuse us,” but it’s also evident in their behavior. In the first clip, they don’t bust down the door of the courthouse; when a civilian briefly opens it, they don’t barge inside. At the end of that encounter, one agent says to the other, “This home is too protected; let’s go see if we can find a vulnerable somewhere else.” Given their reputation for violence in raids, IRL ICE agents are unlikely to give up as easily. 

But that kind of environment allows the training session to be a reasonable intensity for a gamer’s first round of practice responding to ICE, and still be a fun, safe place for people to hang out. According to PitaBreadFace, the main goal of the space wasn’t necessarily to be a specifically anti-ICE training facility, but more so to organize a community and build trust. And this tactical frivolity is a proven method of protest—ask anyone who wore a frog costume to a Portland protest earlier this year.

“ A situation, even though it’s virtual, where you can clearly overwhelm ICE’s numbers and do silly stupid things and work together easily and be connected to each other—it just felt like actually winning,” one gamer said in a clip provided to 404 Media. “It felt like a way to kind of heal some of the burnout.”

A virtual situation also allows players to fire back at ICE in ways that likely wouldn’t be practical in real life. In one clip, for example, two agents are chasing after a vulnerable, yelling, “Hey, stop right there!” 

When they get close enough, the vulnerable drops a Boogie Bomb, an item which forces another player to dance under a disco ball for about three seconds. 

“Oh,” the Boogie-Bombed agent exclaims, before the gamers start laughing. 

The event also had another component. Before the practice ICE raids, gamers went around to practice finding one another, creating groups and building connections. PitaBreadFace described this segment as learning how to “meet your neighbors, know those around you, and establish contact.” A lot of that, according to clips provided to 404 Media, involves doing dance emotes together; in one case, it was a team of about 10 people destroying an in-map mansion and yelling, “Pay your taxes!” 

But it also involved discussions about what community means. In the middle of a “Shout!” dance circle, one gamer said that they first learned the importance of community organizing when protesting the 2017 Muslim ban. 

“ I feel like community taught me that like if enough people came together and there was enough will, anything could happen,” they said. “I remember the first Muslim ban, and just hella people went to the airport, and we were able to petition for people to get released. And they were. It was cool to see that organically happen.”

New Save Collective plans to run more events similar to this one through the end of this year, at which point Fortnite is slated to get rid of the proximity chat mode it uses. PitaBreadFace said the response had been so far overwhelmingly positive.

“ I think gamers represent this constituency of people who are really common-sense,” PitaBreadFace said. “It’s not like they’re even super pro-immigrant. They’re just like, ‘No, this doesn’t make sense. This community member who’s been part of a community for 25 years is being ripped out of his home in the middle of the night. That doesn’t make sense, and we should do something about it.’ We have a lot of people who joined the [Discord] server who are like, ‘I actually don’t know, but I know this is wrong and I’m here to learn and participate.’” 

Hidekazu Araki betreibt in Tokio seit Jahren ein Uhrenfachgeschäft, in dem bereits über 10.000 Uhren von Hand gefertigt hat. ProcessX hat ihn dort besucht und ihm beim Fertigungsprozess über die Schulter geschaut.

(Direktlink)