Noam Shazeer, the engineer who co-wrote the paper that launched the modern AI boom and later returned to Google to lead its Gemini models, has left the company again, this time for OpenAI. TechCrunch reported the hire on Friday, citing people familiar with the move. In the same week, OpenAI brought in Dean Ball, an AI policy adviser in the first Trump administration, according to the report.
The two hires, coming weeks before OpenAI’s expected public listing, show a company filling gaps on both sides of its business. Shazeer supplies technical firepower, a researcher with a record of shipping products at frontier labs. Ball supplies Washington know-how, a voice in the policy debates that will shape how OpenAI’s models are regulated and sold.
Shazeer’s path through the industry is unusual even by the standards of AI’s inner circle. He was an author of the 2017 “Attention Is All You Need” paper, which introduced the transformer architecture that underpins virtually every modern language model. He left Google to found Character.AI, a chatbot company valued in the billions, then returned to Google in 2024 to co-lead the Gemini effort before departing once more.
His arrival at OpenAI reunites him with former colleagues and gives the company a researcher who has built and shipped large models at both of its principal rivals. People familiar with the discussions said OpenAI approached him months ago, offering a role with unusual autonomy and a compensation package tied to the company’s stock ahead of its IPO. The offer combined two things Shazeer is said to value: technical independence and a large equity stake.
Ball’s appointment addresses a different front. He served as an AI policy adviser in the first Trump administration, where he was involved in early efforts to shape the government’s approach to the technology, and he has since been a prominent voice in debates over open-source AI and export controls. His return to government-adjacent work comes at a moment when Washington is deciding how, or whether, to regulate frontier models.
The timing is not accidental. OpenAI is preparing for a listing that would be one of the largest technology offerings in years, and it is doing so under a regulatory spotlight. The company faces questions from antitrust authorities about its partnership with Microsoft, scrutiny of its safety practices, and an active debate in Congress over AI legislation. A policy adviser with White House experience helps the company navigate all three, people familiar with its thinking said.
The hires also signal where the talent war is heading. Google has lost a string of senior researchers to rivals and startups over the past two years, and Shazeer’s departure is among the most prominent. OpenAI, for its part, has been recruiting aggressively ahead of its listing, using equity grants that will vest after the IPO to attract people who might otherwise stay put.
Analysts said the moves are consistent with a company trying to look complete before going public. Investors in AI companies have begun asking harder questions about technical depth and regulatory exposure, and a hiring spree that addresses both is the kind of signal bankers like to see in the weeks before a roadshow.
There are risks. Shazeer’s track record includes building products that delighted users but struggled to find revenue, and his independence has clashed with management at previous employers, people who have worked with him said. Ball, meanwhile, joins a company whose views on open-source models have shifted over time, and his advocacy for openness may sit uneasily with the closed approach OpenAI has favored.
The pre-IPO hiring spree is a familiar pattern in Silicon Valley. Companies about to go public typically lock in senior talent with equity packages that vest after listing, and OpenAI has been unusually aggressive, people familiar with its compensation said. The risk is that the grants create a one-way door: executives who join for the stock may leave once it vests, a churn that has hit other companies after their debuts.
Google has not stood still. The company has promoted internal researchers and made counter-offers to senior staff, and its DeepMind unit continues to publish at the frontier, people familiar with its talent efforts said. But the flow of departures has been steady, and the Gemini leadership team that Shazeer leaves behind is already reorganizing around his exit.
For OpenAI, the hires answer a question investors have pressed since the company’s governance crisis two years ago: can it keep the people who make it valuable? The answer, for now, is that it can still attract them, at a price. Whether it can keep them is the longer test, and the IPO will hand out the first batch of answers.
For now, the message from OpenAI is straightforward: it is staffing up for the next stage, adding the people who can build models and the people who can defend them in Washington. Whether those two groups can work together inside one company, ahead of one of the most scrutinized IPOs in technology history, is the test that awaits.
