Meta’s aggressive hiring push to supercharge its artificial intelligence research has continued unabated, with The Information now reporting that four more researchers have jumped ship from OpenAI to join Mark Zuckerberg’s AI ambitions.

The latest hires — Shengjia Zhao, Jiahui Yu, Shuchao Bi, and Hongyu Ren — follow earlier reported departure of Trapit Bansal, a key contributor to OpenAI’s reasoning model o1, and three others: Lucas Beyer, Alexander Kolesnikov, and Xiaohua Zhai, according to The Wall Street Journal.

These moves mark a sharp escalation in Meta’s quiet but high-stakes war for elite AI talent. Zuckerberg has signaled that Meta’s future is tied to advancing “AI superintelligence,” with plans to embed powerful reasoning models into its consumer products and business tools.

Sources say Meta is offering eye-watering compensation packages, with some top researchers reportedly receiving offers valued at up to $100 million, though Meta’s CTO Andrew Bosworth downplayed the headline figures, saying the actual terms are “more complex.”

A Response to Llama 4’s Underperformance?

The hires come in the wake of April’s launch of Meta’s Llama 4 — a suite of open-weight AI models that, while widely used, reportedly failed to meet internal expectations. Zuckerberg had hoped the release would rival OpenAI’s GPT-4, Anthropic’s Claude, and Google’s Gemini in terms of performance and reasoning capability. Instead, it drew criticism for the model used in Meta’s benchmark testing, with some accusing the company of using a specialized version that didn’t reflect the real-world product.

These setbacks appear to have pushed Meta to redouble its efforts. The company’s AI superintelligence unit — which now includes former OpenAI, DeepMind, and GitHub talent — is being structured similarly to Google DeepMind, operating as a core engine to power future Meta products, including virtual assistants, business agents, and advanced developer tools.

OpenAI Pushes Back

OpenAI CEO Sam Altman, whose company has been under scrutiny from regulators and partners alike in recent months, has downplayed the poaching spree, saying “none of our best people have decided to take him up on that.” But the growing number of high-profile exits tells a more complicated story, especially with researchers involved in OpenAI’s core reasoning and reinforcement learning efforts now switching allegiance.

Tensions between the two companies have reportedly deepened. Meta is said to be courting AI startups for acquisitions or partnerships — including Scale AI, where former CEO Alexandr Wang is now part of Meta’s superintelligence team, and failed talks with Safe Superintelligence, Thinking Machines Lab, and Perplexity.

Why This Matters

AI talent is the new oil, and companies like Meta, OpenAI, Google, and Microsoft are locked in a battle to secure the best minds. Unlike traditional tech recruitment, these roles aren’t just about staffing — they determine who builds the next generation of frontier models.

Meta’s push into reasoning models aims to address a key performance gap. OpenAI, Google, and DeepSeek have all released advanced models that handle complex, multistep problem-solving tasks — and Meta wants to catch up quickly. The success of that effort could define the company’s long-term competitiveness in both the consumer and enterprise AI markets.

In the meantime, OpenAI is reportedly preparing to release a public-facing reasoning model in the coming weeks, which could raise the stakes even further for Meta and others racing to lead in this space.

However, the bottom line is that Meta is building an elite task force to challenge OpenAI and other leading AI companies at their strongest point — reasoning and autonomy.

Microsoft is facing a new lawsuit from a group of prominent authors who accuse the company of using pirated copies of their books to train its Megatron artificial intelligence model.

Filed Tuesday in a federal court in New York, the suit adds to a growing wave of legal challenges from writers, publishers, and news organizations accusing major AI developers of exploiting copyrighted material without consent.

The plaintiffs—who include American Prometheus author Kai Bird, essayist Jia Tolentino, and historian Daniel Okrent—allege that Microsoft trained Megatron using a dataset containing nearly 200,000 pirated books, without notifying or compensating the authors. According to the complaint, the AI model was engineered not just to learn the language but to mimic “the syntax, voice, and themes” of the authors’ works, enabling it to generate text that closely resembles their original expression.

“The Megatron model is not merely informed by our work,” the complaint says, “it is fundamentally built on it—without authorization and without compensation.”

The authors are seeking a court order to block Microsoft’s alleged infringement and statutory damages of up to $150,000 for each work, which could result in millions of dollars in liability if the court rules in their favor. Microsoft did not immediately respond to requests for comment, and the authors’ lawyer declined to speak on the case.

But while the lawsuit appears to be part of a broader campaign by copyright holders to rein in unchecked AI training, recent court rulings suggest Microsoft might actually have the upper hand.

In just the past week, two key decisions have tilted in favor of AI companies. In one, U.S. District Judge William Alsup ruled that Anthropic’s use of copyrighted books to train its Claude AI system fell under “fair use,” calling the process “exceedingly transformative.” Alsup held that training a model to understand and generate language based on large swaths of text was different in purpose from the original works, thereby satisfying one of the primary tests for fair use under U.S. copyright law.

A day earlier, Judge Vince Chhabria dismissed a similar case brought against Meta over its Llama model, stating flatly that “there’s no disputing” the transformative nature of what Meta had built. Chhabria also noted that the authors failed to show the AI system had caused meaningful harm to the market for their books, another key factor in determining fair use.

While neither ruling directly applies to the Microsoft case, legal analysts say the decisions could shape how judges evaluate similar claims moving forward. It is believed that both judges signaled that courts are leaning toward treating AI training as a transformative, fair use of copyrighted material—especially when the output doesn’t directly compete with or replace the original work.

The back-to-back rulings are being hailed as tentative but important victories for companies like Anthropic, Meta, OpenAI, and now, potentially Microsoft. However, the authors’ lawsuit takes a slightly different tack by emphasizing that the company allegedly used pirated copies of their books, which could complicate Microsoft’s defense even if training itself is deemed lawful.

The court may need to examine not just the legality of the AI’s purpose but also the nature of the data it was fed.

The situation underscores the complexities of copyright law in the AI age. Rather than offering clarity, the recent rulings have deepened the complexity of the issue. They suggest that while training on copyrighted material may, in many instances, qualify as fair use, there may still be legal consequences depending on how that material was acquired.

Now, Microsoft’s case has become another test of where the boundaries of fair use end and where infringement begins. Some analysts believe that this issue will require Congress’ or the Supreme Court’s intervention, or this evolving legal gray area is likely to remain unsettled.