Tech Brew Ride Home - Thu. 06/26 – Conflicting AI Legal Rulings

Episode Date: June 26, 2025

The legal rulings on AI are finally coming in. The problem is, they’re contradictory, so we’re not getting any legal clarity yet. Creative Commons but for AI training data. Is DeepSeek’s R2 model being stymied by lack of access to Nvidia chips? And another deep look at the question of: is AI taking jobs at tech companies, right now? Links: Microsoft sued by authors over use of books in AI training (Reuters) Trump Mobile reiterates claims that new phones are 'made in America' (USAToday) Creative Commons debuts CC signals, a framework for an open AI ecosystem (TechCrunch) OpenAI, Microsoft Rift Hinges on How Smart AI Can Get (WSJ) DeepSeek’s Progress Stalled by U.S. Export Controls (The Information) Salesforce CEO Says 30% of Internal Work Is Being Handled by AI (Bloomberg) AI Killed My Job: Tech workers (Blood In The Machine) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the TechMeme right home for Thursday, June 26, 2025. I'm Brian McCullough today. The legal rulings on AI are finally coming in. The problem is they're contradictory, so we're not getting any legal clarity yet. Creative Commons, but for AI training data, is DeepSeaks R2 model being

stymied by lack of access to Nvidia chips. And another deep look at the question of is AI taking jobs at tech companies right now? Here's what you miss today in the world of tech. A group of authors is suing Microsoft in a New York federal court, claiming the company used nearly 200,000 pirated books without permission to train its Megatron AI model, quoting Reuters. Kaya Bird, Gia Tolentino, Daniel Ockrent, and several others allege that Microsoft used pirated digital versions of their books to teach its AI to respond to human prompts. Their lawsuit filed in New York federal court on Tuesday is one of several high-stakes cases brought by authors, news outlets, and other copyright holders against tech companies, including. meta-platforms, Anthropic, and Microsoft-backed OpenAI over alleged misuse of their material in AI training. The complaint against Microsoft came a day after a California federal judge ruled that Anthropic made fair use under U.S. copyright law of author's material to train

its AI systems but may still be liable for pirating their books. It was the first U.S. decision on the legality of using copyrighted materials without permission for generative AI training. The writers alleged in the complaint that Microsoft used a collection of nearly 200,000 pirated books to train Megatron an algorithm that gives text responses to user prompts. The complaint said Microsoft used the pirated dataset to create a, quote, computer model that is not only built on the work of thousands of creators and authors, but also built to generate a wide range of expression that mimics the syntax, voice, and themes of the copyrighted works on which it was trained, end quote. Tech companies have argued that they make fair use of copyrighted material to create new

transformative content and that being forced to pay copyright holders for their work could hamstring the burgeoning AI industry, end quote. Yeah, interestingly, there's more news on this front just today. A different U.S. judge has ruled that Mehta's use of books to train Lama is protected by fair use but says his opinion is more a reflection of the plaintiff's poor arguments. quoting Bloomberg. This ruling does not stand for the proposition that meta's use of copyrighted materials to train its language models is lawful, the judge said. It stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record in support of the right one. Boys Schiller-Flexner-L-L-P, the author's firm, said in a statement that, quote,

despite the undisputed record of meta's historically unprecedented pirating of copyrighted works, the court ruled in meta's favor. We respectfully disagree with that conclusion. A meta-spokesperson said in a statement that fair use of copyright material is a vital legal framework for building this transformative technology. The ruling comes two days after a judge in the same court ruled largely in favor of AI competitor Anthropic, PBC, which was similarly accused of illegally using books to train its models. This judge's opinion, the second ever federal court ruling to weigh in on the use of copyrighted works to train today's most powerful generative AI models, appeared to reject the idea that the industry can use billions of human-made works without seeking permission. It's hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of potentially endless streams of competing works that could significantly harm the market for those books, the judge said. The judge also rejected meta's argument that an adverse ruling would harm AI technology growth by requiring companies to pay for copyrighted training data,

saying that theory, quote, doesn't pass the straight face test. These products are expected to generate billions, even trillions of dollars for the companies that are developed. helping them, the judge said, if using copyrighted works to train the models is as necessary as the company say, they will figure out a way to compensate copyright holders for it. The back-to-back rulings in favor of Meta and Anthropic will have major ramifications for dozens of similar lawsuits across the country against top AI companies, including OpenAI and stability AI, as well as the growing data licensing market for AI models. Although both Meta and Anthropic won their fair use arguments in their respective cases,

their separate rulings reveal a divide over how courts could interpret the legal test. Judge William Alsup, who was overseeing the Anthropic case, ruled on Monday that the company's use of works to train its AI model called Claude is, quote, among the most transformative many of us will see in our lifetimes. Calabria's ruling rebuffed Alps' analysis, saying that the earlier opinion brushed aside the harm that generative AI models will have on the market for the original books. But Chabria said his decision to find in favor of meta came down to the author's lack of any evidence that AI chatbots were undermining their book sales. all the plaintiffs presented is speculation and speculation is insufficient, the judge said, end quote. So yeah, no clarity yet, basically at all. When that Trump phone got announced, what was it, last week, part of the announcement was that it was, quote, made in America, which got a lot of people to say, how exactly,

seeing as how no one else can seemingly make a smartphone entirely in the U.S., well, the Trump organization has apparently replaced the T1 phones made-in-America language on Trump Mobile's site with, quote, brought to life right here in the USA and, quote, proudly American, quoting USA Today. Chris Walker, a Trump mobile spokesperson, dismissed the report saying that T1 phones are proudly being made in America. Speculation to the contrary is simply inaccurate, Walker said in a statement to USA Today. A pre-order message on its website previously referred to the device as made in the USAT1

phone, a June 16th, Getty Images photo illustration shows. In a June 16th interview with Fox Businesses Mornings with Maria, Trump Organization Executive Vice President Eric Trump also alluded to the device being manufactured in the U.S. You're not calling up call centers in Bangladesh, do it right out of St. Louis, Missouri, and you're going to have phones that are made right here in the United States of America, he said, end quote. Creative Commons has debuted CC Signals, a framework that lets data set holders detail how machines can or cannot reuse their content, such as for training, AI models. Quoting TechCrunch. The idea is meant to create a balance between the open nature of the internet and the demand for ever more data to fuel AI.

As Creative Commons explains in a blog post, the continued data extraction underway could erode openness on the internet and could see entities walling off their sites or guarding them with paywalls instead of sharing sharing access to their data. The CC Signals Project, on the other hand, aims to provide a legal and technical solution that would provide a framework for dataset sharing meant to be used between those who control the data and those who use it to try. A.I. Demand is increasing for such a tool as companies grapple with changing their policies

and terms of service to either limit AI training on their data, or explain to what extent they'll utilize users' data for purposes related to AI. For instance, X initially made a change that allowed third parties to train their models on its public data, then later reverse that. Reddit is using its robots. Text file, which is meant to tell automated web crawlers whether they can access its site to restrict bots from scraping its data for training AI. Cloudflare is looking toward a solution that would charge AI bots for scraping as well as tools for confusing them. And open source developers have also built tools to slow down and waste the resources of AI crawlers that didn't respect their no-crawls directives. The CC Signals Project instead proposes a different solution, a set of tools that offers a range of legal enforceability

and that has an ethical weight to them similar to the CC licenses that today cover billions of openly licensed creative works online. The project is only now beginning to take shape. Early designs have been published. on the CC website and GitHub page. The organization is actively seeking public feedback ahead of its plans for an alpha launch early test in November 2025. It will also host a series of town halls for feedback and questions, end quote. Keeping abreast of the OpenAI and Microsoft saga, sources say Microsoft is pushing to remove the AGI clause from its Open AI contract, which lets Open AI limit Microsoft's access to its IP

once its systems achieve artificial general intelligence. quoting the journal. Many AI experts see AGI as the point at which generative AI systems achieve human-like intelligence, but Open AI and Microsoft are at odds over the issue. Open AI executives, including Sam Altman, believe they are close to being able to declare that their AI tools have achieved the AGI level of proficiency, according to people familiar with the matter. Microsoft chief executive Sachin Andella has expressed skepticism that reaching such a benchmark is even possible. Their disagreement mirrors a debate among Silicon Valley's elite about just how sophisticated cutting-edge tools can become. In recent negotiations, Microsoft has indicated

it is willing to accept an equity stake of about 35% in the new for-profit Open AI. People familiar with the matter said, the AGI divide looms over the discussions between the two companies. Altman has said AGI, which his company defines as highly autonomous systems that outperform humans at most economically valuable work, is just around the corner. Microsoft's Nadella has been more critical. Us self-claiming some AGI milestone, that's just nonsensical benchmark hacking, He said on a popular tech podcast in February, the comments surprised some OpenAI officials who said Nadella was once described by Altman as an AGI believer. Open AGI executives have discussed the possibility of declaring AGI through an AI coding agent that exceeds the capabilities of an advanced human programmer, people familiar with the matter said. The contract only requires that Open AIs board declare AGI in good faith, though Microsoft could easily sue the company in response, risking a drawn-out legal battle.

Open AI can also declare a higher tier of AGI called Sufficient AGI when its AI systems are financially capable of paying Microsoft the future profits to which it is entitled. Microsoft has to sign off on sufficient AGI at which point Open AI would get the right to sell or license all of its technology to other cloud providers. Microsoft is restricted from developing AGI on its own under the company's contract, which runs through 2030, according to people familiar with the matter. It is OpenAI's largest outside shareholder and has the ability to veto any major funding decision that dilutes its equity. Some Microsoft executives objected to including the AGI clause when the

partnership was first negotiated in 2019, believing that it was arbitrary and unenforceable. But the company was so behind on AI at the time that Nadella agreed to it, according to people familiar with the matter. Microsoft is hoping to remove the clause entirely from the contract as part of its recent negotiations or secure exclusive access to OpenAI's IP, even after AGI is declared, end quote. The information has sources saying that Deep Seeks' highly anticipated R2 model faces delays due to a shortage of Nvidia server chips in China exacerbated by the U.S. ban on NVIDIA's H20 chips. Quote, DeepSeek owned by hedge fund firm High Flyer Capital Management, hasn't yet determined

the exact timing of R2's release, according to two people with knowledge of the situation. While Deep Seek engineers have been working intensively on R2 over the past several months, CEO Liang Wen Feng, isn't satisfied with the new model's performance, according to two people. The company's engineers are working on refining R2 until Liang gives the green light for release, the people said. If Deepseeks R2 outperforms existing open source models when released, demand for it will overwhelm cloud providers in China, which are already grappling with the dearth of Nvidia chips, employees at the cloud company said. Among cloud customers currently using R1, the majority are running the model with Nvidia's H20 chips, they said. R1, like other Deepseek models, is trained on Nvidia's hardware and software and performs best when running on Nvidia chips.

The model requires less powerful chips to run than LLMs developed by U.S. and most other Chinese tech firms. This made R1 and H20 the perfect fit for each other. The April ban on H20 chips has dashed cloud firms' hopes of stockpiling more H20s, however. Although DeepSeek hasn't determined when it will release R2, it has been in close communications with some Chinese cloud companies providing them with technical specifications to guide their plans for hosting and distributing the model from their servers, according to the cloud company's employees. The plans include lining up enough H20 chips for customers who want to use the full-size version of R2.

For now, the best option is to rely on the existing stockpile of H20 chips in China, which could limit to what extent businesses can use R2. Companies outside China could use R2 more easily if they don't face the same chip restrictions, end quote. Finally today from the, is AI taking all the jobs file? Mark Benioff says, quote, AI is doing 30 to 50% of the work at Salesforce now, end quote, including software engineering and customer service. Quoting Bloomberg. The San Francisco-based software company is also selling an AI product that promises to handle tasks such as customer service without human supervision. Benioff said the tool has reached about 93% accuracy, including for large customers such as Walt Disney.

All of us have to get our head around this idea that AI can do things that before we were doing, Beniof said. we can move on to do higher value work, end quote. Meanwhile, though, at the Blood in the Machine substack, workers at Google, TikTok, Adobe, Dropbox, CrowdStrike, and other tech firms recounted how managers have recently been using AI to justify firing them or speeding up their work. There are several different stories in this piece. It's a very long one. But I'm going to just summarize from one. A current CrowdStrike employee said that, I work at CrowdStrike. I can't share identifying details, but I can't say this. Morale is at an all-time low. On May 7th, Crowdstrike laid off 500 employees, not underperformers, but many recent hires.

Leadership framed it as a strategic realignment, aiming to double down on our highest impact opportunities, in the words of CEO George Kurtz. But make no mistake, this was an AI-driven layoff. Generative AI is infiltrating every corner of the company, internal chat, note-taking, triage, engineering, even customer support. Employees are now expected to use new AI tools weekly, often replacing the live services our customers paid for. Even Quality Assurance now penalizes people for not running their work through AI for basic grammar. CrowdStrike is doing extremely well financially. CRWD stock is near a 52-week high, and yet, while the company celebrates its profits and AI integrations, 500 colleagues, including recent grads who uprooted their lives to move to Texas, are out of work.

Those of us who remain are under pressure to do more with less, longer hours, heavier workloads, no extra compensation. Meanwhile, many of the Gen A.I tools we've been given are buggy or just wrong. Some have even fed misinformation to customers damaging trust and credibility. I won't claim Gen AI has directly increased the per-person workload, but it certainly hasn't reduced it. For many, it simply created more complexity, not less. The company is proud of its best workplace accolades. We used to be proud, too. Now we just feel expendable.

Nothing more for you today. Talk to you tomorrow.