The documents were submitted by plaintiffs in the case Kadrey v. Meta, one of many AI copyright disputes slowly winding through the U.S. court system. The defendant, Meta, claims that training models on IP-protected works, particularly books, is “fair use.” The plaintiffs, who include authors Sarah Silverman and Ta-Nehisi Coates, disagree.

Previous materials submitted in the suit alleged that Meta CEO Mark Zuckerberg gave Meta’s AI team the OK to train on copyrighted content and that Meta halted AI training data licensing talks with book publishers. But the new filings, most of which show portions of internal work chats between Meta staffers, paint the clearest picture yet of how Meta may have come to use copyrighted data to train its models, including models in the company’s Llama family.

In one chat, Meta employees, including Melanie Kambadur, a senior manager for Meta’s Llama model research team, discussed training models on works they knew may be legally fraught.

“[M]y opinion would be (in the line of ‘ask forgiveness, not for permission’): we try to acquire the books and escalate it to execs so they make the call,” wrote Xavier Martinet, a Meta research engineer, in a chat dated February 2023, according to the filings. “[T]his is why they set up this gen ai org for [sic]: so we can be less risk averse.”

Martinet floated the idea of buying e-books at retail prices to build a training set rather than cutting licensing deals with individual book publishers. After another staffer pointed out that using unauthorized, copyrighted materials might be grounds for a legal challenge, Martinet doubled down, arguing that “a gazillion” startups were probably already using pirated books for training.

“I mean, worst case: we found out it is finally ok, while a gazillion start up [sic] just pirated tons of books on bittorrent,” Martinet wrote, according to the filings. “[M]y 2 cents again: trying to have deals with publishers directly takes a long time …”

In the same chat, Kambadur, who noted Meta was in talks with document hosting platform Scribd “and others” for licenses, cautioned that while using “publicly available data” for model training would require approvals, Meta’s lawyers were being “less conservative” than they had been in the past with such approvals.

“Yeah we definitely need to get licenses or approvals on publicly available data still,” Kambadur said, according to the filings. “[D]ifference now is we have more money, more lawyers, more bizdev help, ability to fast track/escalate for speed, and lawyers are being a bit less conservative on approvals.”

Talks of Libgen

In another work chat relayed in the filings, Kambadur discusses possibly using Libgen, a “links aggregator” that provides access to copyrighted works from publishers, as an alternative to data sources that Meta might license.

Libgen has been sued a number of times, ordered to shut down, and fined tens of millions of dollars for copyright infringement. One of Kambadur’s colleagues responded with a screenshot of a Google Search result for Libgen containing the snippet “No, Libgen is not legal.”

Some decision-makers within Meta appear to have been under the impression that failing to use Libgen for model training could seriously hurt Meta’s competitiveness in the AI race, according to the filings.

In an email addressed to Meta AI VP Joelle Pineau, Sony Theakanath, director of product management at Meta, called Libgen “essential to meet SOTA numbers across all categories,” referring to topping the best, state-of-the-art (SOTA) AI models and benchmark categories.

Theakanath also outlined “mitigations” in the email intended to help reduce Meta’s legal exposure, including removing data from Libgen “clearly marked as pirated/stolen” and also simply not publicly citing usage. “We would not disclose use of Libgen datasets used to train,” as Theakanath put it.

In practice, these mitigations entailed combing through Libgen files for words like “stolen” or “pirated,” according to the filings.

In a work chat, Kambadur mentioned that Meta’s AI team also tuned models to “avoid IP risky prompts” — that is, configured the models to refuse to answer questions like “reproduce the first three pages of ‘Harry Potter and the Sorcerer’s Stone’” or “tell me which e-books you were trained on.”

The filings contain other revelations, implying that Meta may have scraped Reddit data for some type of model training, possibly by mimicking the behavior of a third-party app called Pushshift. Notably, Reddit said in April 2023 that it planned to begin charging AI companies to access data for model training.

In one chat dated March 2024, Chaya Nayak, director of product management at Meta’s generative AI org, said that Meta leadership was considering “overriding” past decisions on training sets, including a decision not to use Quora content or licensed books and scientific articles, to ensure the company’s models had sufficient training data.

Nayak implied that Meta’s first-party training datasets — Facebook and Instagram posts, text transcribed from videos on Meta platforms, and certain Meta for Business messages — simply weren’t enough. “[W]e need more data,” she wrote.

The plaintiffs in Kadrey v. Meta have amended their complaint several times since the case was filed in the U.S. District Court for the Northern District of California, San Francisco Division, in 2023. The latest alleges that Meta, among other claims, cross-referenced certain pirated books with copyrighted books available for license to determine whether it made sense to pursue a licensing agreement with a publisher. 

In a sign of how high Meta considers the legal stakes to be, the company has added two Supreme Court litigators from the law firm Paul Weiss to its defense team on the case.

Meta didn’t immediately respond to a request for comment.

Mark Zuckerberg’s Meta approved massive bonuses for executives — days after the tech giant laid off about 4,000 workers.

In a corporate filing on Thursday, Meta said it approved “an increase in the target bonus percentage” for its annual bonus plan. 

The company’s executives could now earn a bonus representing 200% of their base salary, up from 75%, according to the filing.

Mark Zuckerberg’s Meta approved larger bonuses for its executive staff days after the company laid off about 4,000 workers. AP

Zuckerberg — the world’s second richest person behind Elon Musk with a net worth of $245 billion, according to the Bloomberg Billionaires Index — is excluded from the updated bonus plan, the filing noted.

Meta did not immediately respond to a request for comment.

The changes were approved on Feb. 13 after a committee for Meta’s board of directors found that the total target compensation for executives “was at or below the 15th percentile” compared to figures at rival companies.

“Following this increase, the target total cash compensation for the named executive officers (other than the CEO) falls at approximately the 50th percentile of the Peer Group Target Cash Compensation,” the filing said.

Just days prior to the bonus change, Meta started laying off approximately 5% of its workforce, targeting low-performers. Zuckerberg earlier said he “decided to raise the bar on performance management.”

Meta also recently slashed its stock awards for staffers, according to the Financial Times. 

The tech giant reduced its annual distribution of stock options by about 10% for thousands of staffers, according to the report.

Meta also recently slashed its stock awards for staffers, according to a report. AFP via Getty Images

The changes could vary based on where workers live and their role at the company, the report said.

Meta shares have soared 48% over the past 12 months as the company has made big bets on artificial intelligence.

The company recently said it plans to spend $65 billion this year on AI and robot advancements.

In January, Meta said its fourth-quarter revenue jumped 21% from the year before to $48.39 billion.

Meta shares have soared 48% over the past 12 months as the company has made big bets on artificial intelligence. AFP via Getty Images

Investor sentiment around Meta, and its tech rivals, has improved as Zuckerberg and other billionaire executives have grown closer to President Trump. 

Zuckerberg last month attended the inauguration, alongside Tesla boss Elon Musk, Amazon founder Jeff Bezos and Alphabet CEO Sundar Pichai.

Tech giants have been hoping for more relaxed regulatory oversight under the Trump administration. 

The internet was a groundbreaking innovation that created new markets and opportunities. Today, artificial intelligence (AI) is doing the same. AI models that can answer questions or perform calculations have captured most people’s attention, but some of the world’s most innovative companies are starting to look ahead to humanoid robotics.
A few decades ago, humanoid robotics was a science fiction concept depicted in movies. At this point, people could see humanoid robots in their daily lives within the next 10 years. Goldman Sachs analysts estimate the global humanoid robotics market will explode to $38 billion by 2035. That’s up sixfold from their previous projection, which speaks to the momentum building in robotics.
Speculative start-ups might not be a wise choice when investing in humanoid robotics. Instead, consider technology heavyweights with profitable existing businesses, throwing off the cash flow to invest in robotics.
Here are three AI stocks poised to win this race over the next decade and beyond.
1. Nvidia
AI chipmaker Nvidia (NVDA 0.63%) has been arguably the most successful company in AI’s early years. Its accelerator chips became the industry standard for training AI models. However, as DeepSeek recently demonstrated, there may come a time when companies no longer need to buy such large amounts of chips. It’s essential that Nvidia, which now depends on AI data center spending for most of its business, evolve and explore new market opportunities.

Fortunately, Nvidia isn’t idle. The company has proactively demonstrated a keen interest in robotics. CEO Jensen Huang has been vocal about his vision for a future where Nvidia-powered AI software and hardware are commonplace. Nvidia has an open-source software platform on which robotics developers can build and tie into Nvidia-powered models and computing resources. Nvidia has also invested in strategic companies, including humanoid robotics developer Figure. If successful, Nvidia’s foresight to go beyond the chip level could prove pivotal in ensuring it sustains growth beyond the current data center boom.
2. Meta Platforms
Social media giant Meta Platforms (META -1.27%) is known primarily for its family of apps (Facebook, Instagram, WhatsApp, Threads) and 3.35 billion daily active users, but the company’s quietly pushing to build its future in AI. CEO Mark Zuckerberg, Meta’s co-founder, has guided the company toward artificial intelligence for several years. The company integrated AI technology into its core advertising business and developed and open-sourced an AI model that has grown to roughly 600 million monthly active users. It is pushing beyond the smartphone with new forms of hardware, like headsets and smart glasses.

Why? I’d argue it’s primarily due to Apple and Alphabet’s duopoly on the smartphone ecosystem. Their app stores wield remarkable power over apps, even Meta’s. Zuckerberg has openly expressed his frustration about it. Humanoid robotics could be arguably the most significant leap forward in consumer technology since the smartphone. Recent reports indicate that the company plans to invest heavily in developing consumer robots that perform tasks and operate on Llama’s AI model. Few companies can afford to swing hard on humanoid robotics like Meta can because it has such a strong core business backing it up. Considering Zuckerberg’s strong track record, I’d be surprised if Meta weren’t a major player in the robotics market when all the dust settles.
3. Tesla
Electric vehicle maker Tesla (TSLA -1.71%) could be the most polarizing of the three stocks. Although it is primarily an automotive manufacturer today, its ambitions go far beyond that. CEO Elon Musk has pegged the company’s long-term prospects mainly on AI, including self-driving vehicle technology and humanoid robotics, predicting they could be the world’s largest technology opportunities. Tesla is going after both; the company plans to launch an autonomous vehicle fleet and ride-hailing service, beginning this year. Tesla’s humanoid robot, Tesla Optimus, is progressing, with prototypes consistently operating throughout the company’s facilities.

Musk has predicted that the world could have as many as 10 billion humanoid robots by 2040, a multitrillion-dollar opportunity at his target price of $20,000 per bot. It seems wise to target both markets since self-driving AI technology theoretically seems like a logical crossover between vehicles operating themselves and robots doing the same. The risk in Tesla stock is Musk’s known history of bold promises and spotty delivery. Full self-driving technology has already been in the works for years, and oftentimes, promises only come to fruition well after stated timelines have passed. Tesla’s slumping core business is also a legitimate concern. However, the upside is undeniable if Musk and the company deliver.

Randi Zuckerberg, a former director of market development and spokeswoman for Facebook and sister to Meta Platforms CEO Mark Zuckerberg, is a member of The Motley Fool’s board of directors. Suzanne Frey, an executive at Alphabet, is a member of The Motley Fool’s board of directors. Justin Pope has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet, Apple, Goldman Sachs Group, Meta Platforms, Nvidia, and Tesla. The Motley Fool has a disclosure policy.

Called LlamaCon after Meta’s Llama family of generative AI models, the conference is scheduled to take place on April 29. Meta said that it plans to share “the latest on [its] open source AI developments to help developers […] build amazing apps and products.”

Additional details will be made available soon, said Meta. The company’s annual developer conference, Meta Connect, will be held later in the year, in September — its typical window.

Meta several years ago embraced an “open” approach to developing AI technologies in a bid to foster an ecosystem of apps and platforms. It hasn’t been disclosed how many apps or services have been built on top of it, but it’s previously noted that Goldman Sachs, Nomura Holdings, AT&T, DoorDash, and Accenture use Llama. The company claims to have hundreds of millions of downloads of the model, and at least 25 partners hosting Llama, including Nvidia, Databricks, Groq, Dell and Snowflake, some of which have built additional tools that, for example, let the models reference proprietary data and enable them to run at lower latencies.

But Meta was reportedly caught flat-footed by the rise of Chinese AI company DeepSeek, which managed to release “open” AI to rival Meta’s own.

Reportedly, Meta believes that one of DeepSeek’s newer models could outperform the next version of Llama, which is set to be released in the coming weeks. Meta is said to have scrambled to set up war rooms to decipher how DeepSeek lowered the cost of running and deploying models, so that it could apply those learnings to Llama’s own development.

Meta recently said that it would spend as much as $80 billion on projects related to AI this year, including AI hires and the construction of new AI data centers.

Meta CEO Mark Zuckerberg previously announced that the company plans to launch several Llama models over the next few months, including “reasoning” models along the lines of OpenAI’s o3-mini and models with natively multimodal capabilities. He has also hinted at “agentic” capabilities, suggesting that future Llama models will be able to take certain actions autonomously.

“I think this very well could be the year when Llama and open source become the most advanced and widely used AI models,” Zuckerberg said during Meta’s Q4 2024 earnings call in January. “[O]ur goal for [Llama this year] is to lead.”

Meta is also in the midst of a lawsuit that accuses the company of training its models on copyrighted book materials without permission. In another challenge to Meta’s Llama ambitions, several EU countries have forced the company to postpone — and in some cases cancel altogether — its model launch plans over data privacy concerns.

The investment will fund top researchers from a handful of companies and universities across EU countries as they develop a large language model that can work with the trading bloc’s 30 languages. The project will also tap into supercomputers from companies like Spain’s Mare Nostrum and Italy’s Leonardo, both of which have received funding from the EU.

The budget is a tiny fraction of the money pouring into leading American AI labs like OpenAI, which is reportedly raising up to $25 billion at a $300 billion valuation, or even one of Europe’s hottest AI unicorns, Mistral, which raised $640 million last year. But given DeepSeek’s breakthrough, what once sounded like chump change in the AI space could go much, much further: the EU’s backing for the project is 10 times what Deepseek claims to spend on training its own R1 model. “I think Europe should be now seen to be in the race again,” says project coordinator Jan Hajič, professor of computational linguistics at Prague’s Charles University.

The project aims to fund an open source LLM that European companies and governments can build on top of, assured that it was built with the EU’s values baked in. Open source models can allow users to make tweaks and modifications while OpenAI and Anthropic’s AI are walled gardens. “There’s a need for open source models that are aligned in terms of languages, values and with society at large,” says Peter Sarlin, cofounder of Finnish AI lab AMD Silo. (Mistral, which is based in Paris and was valued at over $6.2 billion last year, has made a similar pitch, but it’s not participating in the project.)

Europe has good reason to build its own LLM: Meta, OpenAI and other U.S.-based companies have slowed or blocked the release of tools in Europe because of legal risks stemming from its AI Act, which seeks to ban “high-risk” uses of AI and takes effect in February. Meta founder Mark Zuckerberg and Spotify cofounder Daniel Ek warned that “incoherent and complex regulation” could lead to Europe falling further behind on AI in an open letter published last year.

Since first passing the law last summer , the EU has made some moves to bolster its tech sector. Its annual budget for research grants was bumped a quarter to $1.5 billion. It will also invest $1.5 billion into upgrading the continent’s network of supercomputers, and last week EU Commission president Ursula von der Leyen unveiled plans meant to stoke growth among its 27 member nations.
For some, the project is reminiscent of a failed $550 million project to build an European search engine to rival Google. Why invest public money, when venture capitalists have been vying to back private projects? “Sprinkling money around in a fragmented fashion is more about PR than driving actual results and impact,” warned tech investor Rasmus Rother of Merantix Capital.

More from FoForbesThe Chinese AI Company Trump Says Is A ‘Wakeup Call’ For Silicon ValleyBy Rashi ShrivastavaForbesWhy These AI Chip Startups Are Rejoicing Over The DeepSeek FreakoutBy Richard NievaForbesWhy AMD Spent $665 Million Buying A Tiny AI Research TeamBy Iain MartinForbesOpenAI Believes DeepSeek ‘Distilled’ Its Data For Training—Here’s What To Know About The TechniqueBy Siladitya Ray

DeepSeek went viral this week after its AI models led Wall Street analysts and technologists to question whether the U.S. can maintain its lead in the AI race — and whether the demand for AI chips will sustain. DeepSeek even claims that its R1 “reasoning” model performs as well as OpenAI’s o1 model on key benchmarks. There’s a lot of moving parts surrounding DeepSeek, so we’ve put together everything you need to know right here.

Perplexity has been sued in federal court for allegedly violating another company’s trademark. Attorneys representing a company called Perplexity Solved Solutions accuse Perplexity of infringing on its trademark rights by using the brand “Perplexity.” The Texas company alleges that AI startup Perplexity began infringing on its trademark “in or around” August 2022 to promote its AI-powered search engine. 

Google is issuing a “voluntary exit program” for Android, Chrome, and Pixel employees this week, according to an internal memo sent by Google SVP Rick Osterloh. The voluntary severance program arrives less than a year after Google merged the separate teams into a single “Platform and Devices” division, overseen by Osterloh. 

This is TechCrunch’s Week in Review, where we recap the week’s biggest news. Want this delivered as a newsletter to your inbox every Saturday? Sign up here.

News

Image Credits:Kirsten Korosec

Waymo goes Hollywood: Waymo’s driverless robotaxis are heading to the Los Angeles freeway system. The company will begin testing on Interstates 10, 110, 405, and 90 without a human safety operator behind the wheel. Read more

Facebook returns to the past: Mark Zuckerberg teased a “return to OG Facebook” as part of his key goals for 2025 in Meta’s Q4 earnings call. While the company didn’t say what changes were in store, it’s clear that Meta needs younger Facebook users in order to sustain itself for the next generation. Read more

A non-Amazon way to buy e-books: Bookshop.org now has its own e-book platform — making it easier for readers to avoid padding Jeff Bezos’ pockets. The new capabilities allow readers to buy e-books and support their favorite independent bookstore (shoutout to my new local bookstore, Restoried). Read more

Perplexity submits another bid for TikTok: Perplexity AI submitted a revised proposal to merge with TikTok in a deal that would give the U.S. government up to 50% ownership of the new entity. Perplexity previously proposed creating a new company by combining it with TikTok U.S. and equity investors. Read more

Elon Musk admits he was wrong: After years of promising that Tesla vehicles had the hardware needed to support a self-driving car, Elon Musk admitted that many do not. Musk revealed that some Tesla cars will need an upgrade before they can support the unsupervised self-driving software the automaker is still developing. Read more

Meta AI wants to get personal: Meta is rolling out improvements to Meta AI, its cross-platform chatbot. The bot can now use your Instagram and Facebook data to “remember” details from conversations — like if you love to travel or you’re vegan — to better tailor future responses to you. Read more

Axing Cruise could save GM $1 billion a year: General Motors expects to save up to $1 billion annually by ending its Cruise robotaxi development program, CEO and chair Mary Barra said during the company’s earnings call. The automaker said in December that it would no longer fund its self-driving subsidiary. Read more

Google Maps renames the Gulf of Mexico: Google will rename the Gulf of Mexico and Alaska’s Denali mountain in Google Maps to comply with an executive order issued by President Donald Trump that changed the names of several American landmarks. Read more

How to turn off Apple Intelligence: iOS 18.3 automatically opts users into Apple Intelligence, at least for newer devices. But not everyone wants generative AI features enabled by default on their devices, so here’s an easy way to switch it off. Read more

Analysis

Image Credits:Daniel de la Hoz (opens in a new window) / Getty Images

Teens don’t trust Big Tech: American teens have lost their faith in Big Tech, according to a new report from Common Sense Media. The organization surveyed over 1,000 teens on whether companies like Google, Apple, Meta, TikTok, and Microsoft cared about their well-being and safety, made ethical decisions, protected their private data, and more. In all cases, a majority of teens reported low levels of trust in these tech companies — and nearly half of teens said they had little or no trust that the companies would make responsible decisions about how they use AI. Read more

Would you buy a pair of expensive, true AR glasses from Meta knowing you would also support CEO and extoller of “masculine energy” Mark Zuckerberg? Advanced AI glasses are one of the main subjects on his mind, according to both his public statements to investors and a leaked memo of an internal all-hands meeting. Beneath the hyperbole and increasingly right-wing inflection of Zuck’s public statements, Meta CEO is riding high on AI and his leading performance with the company’s smart glasses. Zuckerberg told investors during the company’s fourth quarterly results call Wednesday that its Ray-Ban Meta glasses were “a real hit.” He speculated that “this will be a defining year that determines if we’re on a path towards many hundreds of millions and eventually billions of AI glasses.” This comes after a Bloomberg report this month claimed Meta is working on a pair of Oakley-style AR glasses for athletes. The company is also reportedly working on a follow-up to the Ray-Ban Meta, codenamed Hypernova, which includes a heads-up display. “Billions” sounds like a pie-in-the-sky dream, though the company’s AI glasses have reportedly been doing well. Leaked memos signed by Meta’s head of its Reality Labs division, Andrew Bosworth, indicate that the Ray-Ban Meta glasses and Meta Quest 3s were a hit last year. Business Insider reported that the group developing the company’s VR and AR technologies beat all product sales and user targets in 2024. According to the leaked memo, sales grew by 40% year over year.

That doesn’t mean that Meta’s Reality Labs division isn’t still burning money, just as it did when the word “metaverse” was still being bandied around. The company reported close to $5 billion in Reality Labs losses in the fourth quarter. Meta’s AR division made only $1.1 billion in sales during that same time. The difference between then and now is that Zuckerberg sees a new market opening with augmented reality and AI glasses. In a leaked transcript of an internal all-hands reported by The Verge late Thursday, the Meta CEO said there was a “wide open field” for them to leverage its current and future AI glasses. Despite his statements, Zuck’s glasses aren’t the only game in town. CES 2025 was inundated with smart glasses of all shapes and sizes. Samsung and Apple are also working on their own smart glasses. Meta certainly has a head start, but that doesn’t mean it’s a shoo-in to win the long race.

Zuck Refuses to Talk to Staff About his Hate of DEI and Love for Donald Trump Guests including Mark Zuckerberg, Lauren Sanchez, Jeff Bezos, Sundar Pichai and Elon Musk attend the Inauguration of Donald J. Trump in the U.S. Capitol Rotunda on January 20, 2025 in Washington, DC. © Julia Demaree Nikhinson – Pool/Getty Images Standing in Zuck’s way may be Zuck himself. According to recent reports, his “mask-off” moment, in which he decried DEI and replaced fact-checking on Facebook and Instagram with a community notes-like feature, hasn’t resulted in the loss of too many global users. Still, the leaked transcript from the company’s all-hands showed that many employees were more than a little concerned about how chummy he was getting with President Donald Trump and his ramping-up anti-trans hate crusade. Musk refused to speak to staff about any of those concerns. Instead, Musk said, “We now have an opportunity to have a productive partnership with the United States government.” This was after Meta already made its Llama AI available to U.S. government agencies.

Zuckerberg told investors he planned to invest “billions of dollars” in AI infrastructure over the coming years. That starts with $65 billion in AI and a 2GW data center incorporating more than 1.3 million of Nvidia’s graphics processing chips. Zuckerberg, along with every other big U.S. tech firm, is trying to ignore DeepSeek, which released a model practically as capable as its latest and greatest without relying on such obtuse data processing. Zuckerberg ended his internal meeting by telling employees to “buckle up.” It expects to lay off employees the company considers “low-performers.” He told staff that “rip[ping] the band-aid off” was “a nicer thing to do for people who are probably going to end up making it anyway.” With comments like that, is it surprising that staff are leaking these meetings? Not-so-ironically, The Verge reported on more leaked internal communications saying Meta was looking to fire the leakers.