Anthropic has poached such an array of high-profile professors that it has become a punch line in academia. “‘I’m joining Anthropic’ is the new meme right now,” Subbarao Kambhampati, a computer-science professor at Arizona State University (who has not joined Anthropic), told us. This month, the AI company hired the chair of UC Berkeley’s department of electrical engineering and computer science, presumably to help build more capable bots. Perhaps more surprisingly, Anthropic has in recent weeks also picked up a Stanford economist, a theoretical physicist from the University of Maryland, and an analytic philosopher from UT Austin.

AI companies are turning into something like mini-universities in their own right. OpenAI employs top mathematicians and physicists—including one who studies black holes, and another who specializes in string theory. At least three computer-science professors joined Meta’s AI lab in late June. DeepMind, like Anthropic, is home to a crew of philosophers. And Anthropic’s recent job postings indicate an interest in hiring legal scholars and political scientists. It’s unclear exactly how many current and former professors are working at AI companies. Across these four firms, we found more than 80—the majority of whom are computer scientists. Some have left academia entirely; others are still working part-time at a university. That number is likely a significant undercount, because we don’t have access to internal data; it also doesn’t include the many professors who have started their own companies, those at other AI start-ups, and those working with the industry in a less formal capacity.

Particularly for AI researchers, these companies have a strong gravitational pull. “Much of the important research is being done in industry now,” Humphrey Shi, a computer-science professor at Georgia Tech who joined Nvidia as a vice president last fall, told us. As he sees it, “If you want to do something that really, truly matters, you probably want to join one of those entities.” Tech firms are making offers—including very enticing salaries—that are hard for academics to refuse. In the process, research that previously would have happened in the open is getting locked up behind closed doors.

For decades, universities were the center of AI research. The field itself officially began at a gathering of researchers at Dartmouth in 1956, and federal funding provided much of the field’s early support. In the early 2010s, Silicon Valley executives began to take AI’s commercial potential more seriously, and set out to hire the best researchers. In 2013, Google paid $44 million to acquire a start-up run by a trio of AI researchers from the University of Toronto. Facebook then hired Yann LeCun, an NYU professor, to establish the company’s AI-research lab; Uber poached some 40 Carnegie Mellon researchers to work on driverless cars. But for the most part, even as more work was being done inside of private companies, many of the newly hired academics at tech companies retained their professorships and established a culture of open research. “Researchers will be strongly encouraged to publish their work,” OpenAI wrote in its founding announcement. (Note the organization’s name.) This norm helped lead to the current AI boom: In 2017, scientists at Google published a research paper that was immediately of interest to OpenAI. Google’s innovation, called the “transformer,” is what the T in ChatGPT stands for.

As AI has taken center stage, Silicon Valley has intensified its efforts to recruit star researchers—and looked beyond computer-science departments. Philosophers help train tech companies’ bots to better interact with humans, and economists study the labor-market implications of AI. Compensation is only part of the draw. The current era of AI research requires massive amounts of computing power. Universities have only a fraction of the resources that Silicon Valley can provide, and the chasm has widened as the Trump administration has cut back on scientific funding. Anca Dragan, a UC Berkeley computer scientist who heads DeepMind’s AI-safety-and-alignment department, wrote to us that she was partly motivated to take the job to acquire “the data, compute, and budget access to make progress on safety at the frontier.” Some professors who remain in academia are also forming partnerships with frontier labs or starting their own AI companies, in part so that they can pursue their research without resource constraints. And in some cases, would-be star Ph.D. students are dropping out of their graduate programs—or skipping them altogether—to pursue careers in AI instead.

[Read: Someone finally wants to hire philosophers]

Although plenty of research is still happening within universities, many professors told us, the result is a flywheel: As more academics get sucked up by industry, the center of AI research moves further from academia, thus increasing the incentive for remaining researchers to leave. The exodus could have some advantages. Many of the professors currently at the labs are on temporary leave; some will likely return to academia full-time. When they do so, they’ll bring new knowledge about frontier research to their institutions. (Dragan is currently preparing to teach a small Ph.D. seminar on AI safety.) Scientists at AI labs might also innovate at a faster pace. After all, they don’t have to worry about applying for grants or waiting years for an article to go through peer review. “If some of the best research is being done in these companies, then I kind of want colleagues and faculty members to be there,” Chris Gregg, a Stanford computer scientist, told us. “You go where the best research is being done, and if that happens to be at a company, so be it.”

But when academics decamp to AI labs or spend more of their time working on their own companies, universities are left with fewer professors to teach the next generation of researchers. (Gregg said that students occasionally ask why courses are no longer offered. The answer is sometimes that the professor is on leave at an AI company.) When star professors leave universities, it becomes harder for students to do work that helps them stand out to prospective employers, Shi said. To differentiate themselves from other applicants, it helps to have worked with mentors who are doing consequential or cutting-edge research. Jennifer Chayes, the dean of the College of Computing, Data Science, and Society at UC Berkeley, told us she fears that as research concentrates in labs with proprietary models, it will be harder for people outside the labs to use AI models to advance science. “Computer-science departments at universities will survive this,” she said. “I don’t know if our innovation economy will.”

As the AI race intensifies, companies now decline to release much of their research for fear of giving away their competitive advantage. Both Gregg and Chayes said they hear from researchers who have gone to AI labs that they can’t publish the work they want to. “The top few AI companies share some things that are at the cutting edge, but it’s a very narrow slice of the potential AI literature that could exist,” Nathan Lambert, an independent AI researcher, told us. According to one study, when AI experts permanently transition from universities to companies, they publish roughly 65 percent fewer papers a year. Papers that do get published tend to emphasize progress on safety efforts and not frontier AI capabilities; corporate research can help fuel marketing hype. And with competition among AI companies escalating, it’s not only the research that is getting locked down, but AI models too. Last month, when Anthropic released a new powerful model called Fable, the company announced that it would invisibly degrade the model’s ability to do certain AI research. The move was positioned as a safety decision, but the resulting outcry from academia was strong. (Anthropic apologized and changed the policy.)

[Read: There’s never been a better time to study computer science]

Academics inside of the tech companies are more sanguine. Shi, the Georgia Tech professor who works at Nvidia, said the company is committed to open research, which is partly why he took the job. Meanwhile, Dragan, the Berkeley computer scientist at DeepMind, explained that publishing isn’t everything. “I find a much bigger priority than publishing is engaging in policy and regulation,” she wrote to us. We also reached out to spokespeople at OpenAI, Meta, and Anthropic for comment. Anthropic declined to comment; OpenAI and Meta did not respond.

The implications are potentially profound, not just for AI research, but for science more generally. In Silicon Valley, there’s a popular view that leading AI labs will end up as the primary drivers of science. “AGI has the potential to be the ultimate tool for advancing science and medicine,” DeepMind CEO Demis Hassabis wrote last week. Hassabis has previously said that an inspiration for his company is Bell Labs, the R&D arm of AT&T, which employed some 1,200 Ph.D.s at its height in the 1960s. Scientists at the lab earned at least 10 Nobel Prizes, and invented the transistor and the modern solar cell. Already, DeepMind has done Nobel Prize–winning research on protein-structure prediction, and the top models have already demonstrated impressive capabilities in mathematics research. On Sunday, a mathematician who works at Anthropic posted that he had used the company’s Fable model during the World Cup final to resolve a nearly 90-year-old mathematical problem.

But if scientific talent and computing power becomes concentrated inside private firms, Silicon Valley could also end up as a gatekeeper of science. Some are raising concerns: A group of academics recently published a declaration warning about “the increasing involvement of technology companies in mathematical research.” If left unchecked, they argued, the incursion of tech companies into research could affect “the scope and depth of mathematical research itself.” Instead of accelerating science, as the leaders of AI companies claim they will, they might end up suffocating it.

In an impressive challenge to the idea that no publicity is bad publicity, Anthropic is now promoting itself with images of disaster, death, and human suffering. The AI company’s latest ad, which debuted during a World Cup match last week, is a perplexing 90-second artifact. The commercial opens with footage of a house engulfed in flames and the crackling sound of burning wood. From there, a series of dark images—rows of gravestones in a cemetery, a homeless couple sleeping on the street, laborers working in a West African granite quarry—flashes across the screen. These stills are accompanied by a series of voice-overs: “Can AI be trusted?” “Who’s going to hit the brakes if we need to?” “How do we really ensure that what we’re aiming to achieve really does benefit the majority of people?” The questions are left unanswered.

Halfway through, the ad moves from dystopian film trailer to feel-good corporate ad. We see a photo of a mother sitting beside her daughter as a voice speculates that AI will allow her to be a better mom. Later, a medical worker uses a laptop covered in Anthropic stickers. “Maybe it’ll cure some great things,” a man says. A nature montage plays—there’s a humpback whale, grassy hills, and people hugging.

Finally, text flashes on the screen: “There’s hope in hard questions.” In the background, we see the ocean. And then Anthropic’s tagline: “Keep thinking.”

The commercial is tone-deaf. It’s one thing to mention the possibility of hitting the brakes on AI progress; it’s another to overlay the question on an image of dead Americans. The video is also just confusing. Does Anthropic think that AI is going to set the world on fire and leave us jobless, or that its chatbot, Claude, will lead us to a more human future where we spend our days watching whales by the sea? The company’s answer seems to be: We don’t know. The video has been mocked, obviously. Some people likened it to hypothetical ads by oil companies professing concerns about the environment. The head of one tech-media company posted, “I actually like the tagline but we don’t need the shots of *checks notes* burning houses and graves!” Meanwhile, OpenAI’s Sam Altman—who has a self-interest in undercutting Anthropic—chimed in to call the ad “satire.”

This is not the first time that a company has botched its advertising around AI: OpenAI has had its own share of embarrassing ads. By now, it seems like a rite of passage. Two years ago, Apple ran a commercial depicting musical instruments and art supplies being crushed by a hydraulic press. The ad was supposed to show off just how much creativity could be compacted into an iPad, but instead it seemed to symbolize generative AI’s bludgeoning of the arts. Apple subsequently apologized. During the Olympics that summer, Google ran an ad in which a young girl wants to write fan mail to the track star Sydney McLaughlin-Levrone—so her dad fires up Gemini and asks the bot for help. The ad was highly unpopular, and Google pulled it from TV. If anything, Anthropic has had a comparatively good run until now. During this year’s Super Bowl, the company aired a commercial dunking on OpenAI for putting ads in user conversations, which even Altman said made him laugh (despite his qualms with how the ad presented OpenAI).

[Read: Watch Apple trash-compact human culture]

For all the criticism, Anthropic’s latest commercial captures the company’s ambivalence about its own existence. Here is a leading AI company worth nearly $1 trillion arguing that the very technology it is racing to commercialize may well imperil the world. “We founded Anthropic because we believe the impact of AI might be comparable to that of the industrial and scientific revolutions, but we aren’t confident it will go well,” the company wrote in 2023. In recent years, Anthropic has made a show of positioning itself as the ethical AI company. Earlier this year, it attracted the ire of the Trump administration when it pushed back against the Pentagon’s demands for unrestricted access to its AI models over concerns that the technology could be misused.

The new advertisement is part of a broader campaign Anthropic is running to address the “hard questions” raised by AI’s progress. In an email, an Anthropic spokesperson pointed me to a series of blog posts about the initiative: This spring, the company traveled across the country to better understand the public’s views about AI. In the video’s description on YouTube, Anthropic says that every question in the ad comes from a person that the company spoke with. (The spokesperson declined to comment further.)

The idea that Anthropic is worried about AI’s negative impacts and wants to hear from the public is certainly better than the alternative. The trouble, though, is that as Anthropic has matured, the company’s constant messaging about its anxiety over the future of AI has become less palatable and, occasionally, downright maddening. “Anthropic is at war with itself,” my colleague Matteo Wong wrote in January. Since then, the company’s valuation has soared. As it develops more advanced AI models and races toward a public offering, the contradictions are becoming more difficult to square. However sincere Anthropic may be—Wong noted after several visits to the company that its employees can be “almost pathologically earnest”—it’s no surprise that people are wary of the world’s most valuable private company asking the “hard questions” about AI.

[Read: Anthropic is at war with itself]

The same tension is at play across the industry. OpenAI was founded as a nonprofit with a mission to “ensure that artificial general intelligence benefits all of humanity.” In the years since ChatGPT was first released, the company seemingly has prioritized financial gain instead. By now, it feels pointless for Anthropic and the rest of the industry to simply pose questions about AI ethics and safety. These questions are already consuming the public’s imagination amid a mounting AI backlash. What matters is the answers. And on that front, AI companies have less to offer. Perhaps the commercial’s tagline provides a helpful clue: For now, Anthropic will keep thinking.