It’s extremely good that Claude’s source-code leaked
If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
Anthropic's developers made an extremely basic configuration error, and as a result, the source-code for Claude Code – the company's flagship coding assistant product – has leaked and is being eagerly analyzed by many parties:
https://news.ycombinator.com/item?id=47586778
In response, Anthropic is flooding the internet with "takedown notices." These are a special kind of copyright-based censorship demand established by section 512 of the 1998 Digital Millennium Copyright Act (DMCA 512), allowing for the removal of material without any kind of evidence, let alone a judicial order:
Copyright is a "strict liability" statute, meaning that you can be punished for violating copyright even if you weren't aware that you had done so. What's more, "intermediaries" – like web hosts, social media platforms, search engines, and even caching servers – can be held liable for the copyright violations their users engage in. The liability is tremendous: the DMCA provides for $150,000 per infringement.
DMCA 512 is meant to offset this strict liability. After all, there's no way for a platform to know whether one of its users is infringing copyright – even if a user uploads a popular song or video, the provider can't know whether they've licensed the work for distribution (or even if they are the creator of that work). A cumbersome system in which users would upload proof that they have such a license wouldn't just be onerous – it would still permit copyright infringement, because there's no way for an intermediary to know whether the distribution license the user provided was genuine.
As a compromise, DMCA 512 absolves intermediaries from liability, if they "expeditiously remove" material upon notice that it infringes someone's copyright. In practice, that means that anyone can send a notice to any intermediary and have anything removed from the internet. The intermediary who receives this notice can choose to ignore it, but if the notice turns out to be genuine, they can end up on the hook for $150,000 per infringement. The intermediary can also choose to allow their user to "counternotify" (dispute the accusation) and can choose to reinstate the material, but they don't have to. Just as an intermediary can't determine whether a user has the rights to the things they post, they also can't tell if the person on the other end of a takedown notice has the right to demand its removal. In practice, this means that a takedown notice, no matter how flimsy, has a very good chance of making something disappear from the internet – forever.
From the outset, DMCA 512 was the go-to tool for corporate censorship, the best way to cover up misdeeds. I first got involved in this back in 2003, when leaked email memos from Diebold's voting machine division revealed that the company knew that its voting machines were wildly insecure, but they were nevertheless selling them to local election boards across America, who were scrambling to replace their mechanical voting machines in the wake of the 2000 Bush v Gore "hanging chad" debacle, which led to Bush stealing the presidency:
The stakes couldn't be higher, in other words. Diebold – whose CEO was an avowed GW Bush partisan who'd promised to "deliver the votes for Bush" – was the country's leading voting machine supplier. The company knew its voting machines were defective, that they frequently crashed and lost their vote counts on election night, and that Diebold technicians were colluding with local electoral officials to secretly "estimate" the lost vote totals so that no one would hold either the official or Diebold responsible for these defective machines:
https://www.salon.com/2003/09/23/bev_harris/
Diebold sent thousands of DMCA 512 takedown notices in an attempt to suppress the leaked memos. Eventually, EFF stepped in to provide pro-bono counsel to the Online Policy Group and ended Diebold's flood:
Diebold wasn't the last company to figure out how to abuse copyright to censor information of high public interest. There's a whole industry of shady "reputation management" companies that collect large sums in exchange for scrubbing the internet of information their clients want removed from the public eye. They specialize in sexual abusers, war criminals, torturers, and fraudsters, and their weapon of choice is the takedown notice. Jeffrey Epstein spent tens of thousands of dollars on "reputation management" services to clean up his online profile:
There are lots of ways to use the takedown system to get true information about your crimes removed from the internet. My favorite is the one employed by Eliminalia, one of the sleazier reputation laundries (even by the industry's dismal standards).
Eliminalia sets up WordPress sites and copies press articles that cast its clients in an unfavorable light to these sites, backdating them so they appear to have been published before the originals. They swap out the bylines for fictitious ones, then send takedowns to Google and other search engines to get the "infringing" stories purged from their search indices. Once the original articles have been rendered invisible to internet searchers, Eliminalia takes down their copy, and the story of their client's war crimes, rapes, or fraud disappears from the public eye:
The takedown system is so tilted in favor of censorship that it takes a massive effort to keep even the smallest piece of information online in the face of a determined adversary. In 2007, the key for AACS (a way of encrypting video for "digital rights management") leaked online. The key was a 16-digit number, the kind of thing you could fit in a crossword puzzle, but the position of the industry consortium that created the key was that this was an illegal integer. They sent hundreds of thousands of takedowns over the number, and it was only the determined action of an army of users that kept the number online:
The shoot-first, ask-questions-never nature of takedown notices makes for fertile ground for scammers of all kinds, but the most ironic takedown ripoffs are the Youtube copystrike blackmailers.
After Viacom sued Youtube in 2007 over copyright infringement, Google launched its own in-house copyright management system, meant to address Viacom's principal grievance in the suit. Viacom was angry that after they had something removed from Youtube, another user could re-upload it, and they'd have to send another takedown, playing Wack-a-Mole with the whole internet. Viacom didn't want a takedown system, they wanted a staydown system, whereby they could supply Google with a list of the works whose copyrights they controlled and then Youtube would prevent anyone from uploading those works.
(This was extremely funny, because Viacom admitted in court that its marketing departments would "rough up" clips of its programming and upload them to Youtube, making them appear to be pirate copies, in a bid to interest Youtube users in Viacom's shows, and sometimes Viacom's lawyers would get confused and send threatening letters to Youtube demanding that these be removed:)
Youtube's notice-and-staydown system is Content ID, an incredibly baroque system that allows copyright holders (and people pretending to be copyright holders) to "claim" video and sound files, and block others from posting them. No one – not even the world's leading copyright experts – can figure out how to use this system to uphold copyright:
However, there is a large cohort of criminals and fraudsters who have mastered Content ID and they use it to blackmail independent artists. You see, Content ID implements a "three strikes" policy: if you are accused of three acts of copyright infringement, Youtube permanently deletes your videos and bars you from the platform. For performers who rely on Youtube to earn their living – whether through ad-revenues or sponsorships or as a promotional vehicle to sell merchandise, recordings and tickets – the "copystrike" is an existential risk.
Enter the fraudster. A fraudster can set up multiple burner Youtube accounts and file spurious copyright complaints against a creator (usually a musician). After two of these copystrikes are accepted and the performer is just one strike away from losing their livelihood, the fraudster contacts the performer and demands blackmail money to rescind the complaints, threatening to file that final strike and put the performer out of business:
The fact that copyright – nominally a system intended to protect creative workers – is weaponized against the people it is meant to serve is ironic, but it's not unusual. Copyright law has been primarily shaped by creators' bosses – media companies like Viacom – who brandish "starving artists" as a reason to enact policies that ultimately benefit capital at the expense of labor.
That was what inspired Rebecca Giblin and I to write our 2022 book Chokepoint Capitalism: how is it that copyright has expanded in every way for 40 years (longer duration, wider scope, higher penalties), resulting in media companies that are more profitable than ever, with higher gross and net revenues, even as creative workers have grown poorer, both in total compensation and in the share of the profits they generate?
https://chokepointcapitalism.com/
The first half of Chokepoint Capitalism is a series of case studies that dissect the frauds and scams that both media and tech companies use to steal from creative workers. The second half are a series of "shovel-ready" policy proposals for new laws and rules that would actually put money in artists' pockets. Some of these policy prescriptions are copyright-related, but not all of them.
For example, we have a chapter on how the Hollywood "guild" system (which allows unionized workers to bargain with all the studios at once) has been a powerful antidote to corporate power. This is called "sectoral bargaining" and it's been illegal since 1947's Taft-Hartley Act, but the Hollywood guilds were grandfathered in. When we wrote about the power of sectoral bargaining, it was in reference to the Writers Guild's incredible triumph over the four giant talent agencies, who'd invented a scam that inverted the traditional revenue split between writer and agent, so the agencies were taking in 90% and the writers were getting just 10%:
Notably, the writers strike was a labor action, not a copyright action. The writers weren't demanding a new copyright that would allow them to control whether their work could be used to train an AI. They struck for the right not to have their wages eroded by AI – to have the right to use (or not use) AI, as they saw fit, without risking their livelihoods.
Right now, many media companies are demanding a new copyright that would allow them to control AI training, and many creative workers have joined in this call. The media companies aren't arguing against infringing uses of AI models – they're arguing that the mere creation of such a model infringes copyright. They claim that making a transient copy of a work, analyzing that work, and publishing that analysis is a copyright infringement:
Here's a good rule of thumb: any time your boss demands a new rule, you should be very skeptical about whether that rule will benefit you. It's clear that the media companies that have sued the AI giants aren't "anti-AI." They don't want to prevent AI from replacing creative workers – they just want to control how that happens.
When Disney and Universal sue Midjourney, it's not to prevent AI models from being trained on their catalogs and used to pauperize the workers whose work is in those catalogs. What these companies want is to be paid a license fee for access to their catalogs, and then they want the resulting models to be exclusive to them, and not available to competitors:
These companies are violently allergic to paying creative workers. Disney takes the position that when it buys a company like Lucasfilm, it secures the right to publish the works Lucasfilm commissioned, but not the obligation to pay the royalties that Lucasfilm owes when those works are sold:
As Theresa Nielsen Hayden quipped during the Napster Wars: "Just because you're on their side, it doesn't mean they're on your side." If these companies manage to get copyright law expanded to restrict scraping, analysis, and publication of factual information, they won't use those new powers to increase creators' pay – they'll use them the same way they've used every new copyright created in the past 40 years, to make themselves richer at the expense of artists:
The Claude Code leak is full of fascinating information about a tool that – like Diebold's voting machines – is at the very center of the most important policy debates of our time. Here's just one example: Claude is almost certainly implicated in the US missile that murdered a building full of little girls in Iran last month:
Of course I see the irony. Anthropic has taken an extremely aggressive posture on copyright's "limitations and exceptions," arguing that it can train its models on any information it can find, and that it can knowingly download massive troves of infringing works for that purpose. It's darkly hilarious to see the company firehosing copyright complaints by the thousands in order to prevent the dissemination, dissection and discussion of the source-code that leaked due to the company's gross incompetence:
But what's objectionable about Anthropic – and the AI sector – isn't copyright. The thing that makes these companies disgusting is their gleeful, fraudulent trumpeting about how their products will destroy the livelihoods of every kind of worker:
If the media bosses get their way, and manage to make it even more illegal – and practically harder – to host, discuss, and publish facts about copyrighted works, then leaks like the Claude Code disclosures will never see the light of day. It's only because of decades of hard-fought battles to push back on this nonsense that we are able to identify and learn about the defects in Claude Code that are revealed by this source-code leak.
I'm angry about the AI industry, but not because of copyright. I'm angry at them for the reasons Cat Valente articulated so well in her "Blood Money" essay:
They took the books I wrote for children and used them to make it possible for children to not bother with reading ever again. They took the books I wrote about love to create chatbots that isolate people and prevent them from finding human love in the real world, that make it difficult for them to even stand real love, which is not always agreeable, not always positive, not always focused on end-user engagement. They took the books I wrote about hope and glitter in the face of despair and oppression and used it to make a Despair-and-Oppression generator.
These goals are entirely compatible with copyright. The New York Times is suing over AI – and they're licensing their writers' words to train an AI model:
If we creative workers are going to pour our scarce resources into getting a new policy to address the threats that our bosses – and the AI companies they are morally and temperamentally indistinguishable from – represent to our livelihoods, then let that new policy be a renewed sectoral bargaining right for every worker. It was sectoral bargaining (a collective, solidaristic right) and not copyright (an individual, commercial right) that saw off AI in the Hollywood writers' strike.
Copyright positions the creative worker as a small business – an LLC with an MFA – bargaining B2B with another firm. To the extent that copyright helps us, it is largely incidental. Sure, we were able to file for a few thousand bucks per book that Anthropic downloaded from a pirate site to train its models on. But Anthropic doesn't have to use a shadow library to get those books – it can just pay our bosses to get them.
It's great that Claude Code's source is online. It's great that we have the ability to pore over, analyze and criticize this code, which has become so consequential in so many ways. It's great the copyright is weak enough that this is possible (for now).
Expanding copyright will gain little for creative workers, except for a new reason to be angry about how our audiences experience our work. Expanding labor rights will gain much, for every worker, including our audiences. It's an idea that our bosses – and AI hucksters – hate with every fiber of their beings.
And let's not forget the next part of the saga, where people are fighting back Anthropic by pasting their leaked code into an LLM and converting it back into something else (in this case a "reimplementation" in Python and Rust), which is afterwards technically not DMCA-able.
If they manage to argue that this is still copyright infrigment they have inadvertently implicated themselves as well as they've been doing this themselves. If they can't then the conversion of the leaked code can however remain in the public domain.
Not to mention that Anthropic is/was the subject of multiple copyright infringement lawsuits, and their entire business case (along with the others) is based on stolen/pirated data.
The houses of cards are falling all around us now. What will happen next? I saw Putin tonight in the parking lot of a supermarket, gathering up shopping carts.
I didn't set out to make multiple posts about the Burgerville CEO, but I've always wondered how they managed to scrub virtually all record of their bizarre "the purpose of Burgerville is to bind workers' souls to their bodies" speech from the Internet to such a degree that I thought I had literally lost my mind until I found a single Tweet referencing it.
Either that poster and I are the only survivors of the collapsed paperwork universe where that happened or (slightly more likely) they paid a fortune to wipe all reporting of it from the Internet using the techniques cited here.
Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions
Free to watch • No registration required • HD streaming
My Dad had a press photo of this when I was a kid, which he kept in a frame. Seeing this picture in the wild is like running across a picture of my brother.
Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions
Free to watch • No registration required • HD streaming
Dwayne Johnson revealed he's losing weight to play a 70-year-old man whose best friend is a chicken in Benny Safdie's "Lizard Music."
"'Benny pitched me this after,' Johnson said during a career retrospective talk at the Toronto Film Festival on Monday where 'The Smashing Machine' will screen. 'And after about 45 minutes, this pitch ended and I said, ‘I am your Chicken Man.''"
Learning that Benny Safdie is making an adaptation of a Daniel Pinkwater book is kind of blowing my mind. As far as I know, the only adaptation of his that ever actually got made is the made for TV version of The Hoboken Chicken Emergency. I think Disney was considering adapting Borgel at one point, but got cold feet.
being anti ai is making me feel like in going insane. "you asked for thoughts about your characters backstory and i put it into chat gpt for ideas". studies have proven its making people dumber. "i asked ai to generate this meal plan". its causing water shortages where its data centers are built. "ill generate some pictures for the dnd campaign". its spreading misinformation. "meta, generate an image of this guy doing something stupid". its trained off stolen images, writing, video, audio. "i was talking with my snapchat ai-" theres no way to verify what its doing with the information it collects. "youtube is impletmenting ai based age verification". my work has an entire graphics media department and has still put ai generated motivational posters up everywhere. ai playlists. ai facial verification. google ai microsoft ai meta ai snapchat ai. everyone treats it as a novelty. every treats it as a mandatory part of life. am i the only one who sees it? am i paranoid? am i going insane? jesus fucking christ. if i have to hear one more "well at least-" "but it does-" "but you can-" im about to lose it. i shouldnt have to jump through hoops to avoid the evil machine. have you no principles? no goddamn spine? am i the weird one here?
Having the head of my division at work tell everyone they should be using AI every day made me lose my mind a bit. I mean, I understand why the boss would want workers to outsource their labor to a machine, but why would we comply?
The small amount of enjoyment I get out of my job is writing some code that actually solves a problem. I don't get wanting to farm that part out to AI so that I am then left with the "go into someone else's code to figure out what went wrong" part of the job.
Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions
Free to watch • No registration required • HD streaming
I love that some executive in 1985 — a full nine years after Convoy was a hit — clearly asked Hoyt Curtin if he could come up with a C.W. McCall knockoff for their bizarre racist magic bear cartoon
Shout out to the production designers who have been putting Cub posters and records in things for the last thirty years. I had always assumed that it was just one person in Vancouver, but the Time Bandits appearance (shot in New Zealand) has caused me to suspect it's some kind of industry-wide conspiracy.
Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.
✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions
Free to watch • No registration required • HD streaming