Noam Shazeer, the Transformer co-author, is leaving Google for OpenAI

Noam Shazeer, the Google engineering vice president who co-led Gemini and co-authored the 2017 "Attention Is All You Need" paper, is leaving Google for OpenAI. The move, reported by CNBC and Reuters on June 18, 2026, lands just days before OpenAI staff were told to expect a public offering within a year. It is the second high-profile hire OpenAI has disclosed inside a week.

What Shazeer's move signals for OpenAI

For OpenAI, this is not just a senior hire. It is a public flex aimed at two audiences at once: technical talent who watched Shazeer quietly re-enter Google two years ago, and the institutional investors OpenAI will need to woo during the IPO roadshow it has begun preparing for.

The technical case is straightforward. Shazeer is one of a small handful of researchers whose fingerprints are visible on every modern large language model. As a Google Brain scientist in 2017, he co-wrote "Attention Is All You Need" with seven colleagues, introducing the Transformer architecture that became the foundation for GPT, Claude, Gemini, Llama, and most of the rest of the current model zoo. He later led the LaMDA project, then spent a year outside Google running Character.AI, a consumer chatbot startup that Google rehired him to lead in a roughly $2.7 billion deal in 2024. He returned to Mountain View as a vice president of engineering and a co-lead of Gemini.

At OpenAI, his role has not been publicly detailed. TechCrunch and Computerworld both note that the company declined to specify what he will work on. That silence is itself a signal. Senior researchers of Shazeer's caliber are usually hired into a specific research agenda or product track, and the deliberate vagueness suggests OpenAI is keeping its roadmap opaque on purpose while the IPO paperwork moves through the SEC.

The investor case is even more direct. The market currently values OpenAI at a price that assumes a credible path to leadership across foundation models, agent infrastructure, and policy access. Hiring one of the most-cited researchers alive, weeks before S-1 filings, is a way to underwrite all three pillars at once.

Shazeer is not the only OpenAI hire this week

The other half of the story is Dean Ball, who spent part of last year inside the Trump White House helping draft America's AI Action Plan before stepping back to the Foundation for American Innovation. According to his own blog post and TechCrunch's reporting, Ball will start at OpenAI on July 6 and will run a small team called Strategic Futures, reporting directly to Chief Strategy Officer Jason Kwon.

The team's remit, as Ball described it, covers four areas: catastrophic risk, recursive self-improvement, labor market impact, and the relationship between frontier labs, governments, and society. He added that the unit will cover both public-facing policy and internal governance. That last bit is the part that matters most for the IPO narrative. Investors who buy into a frontier AI lab are now buying into a regulatory exposure that did not exist two years ago. Having a former White House AI policy architect inside the building, with a direct reporting line to the chief strategy officer, is a way to make that exposure legible to underwriters.

The Ball hire also lands in a specific Washington moment. Late the previous week, the Trump administration had moved to restrict the export of Anthropic's most powerful models, forcing Anthropic to take some of them down. OpenAI is positioning itself as the lab that talks to the White House rather than the one that gets acted on, and Ball is one of the people who actually wrote the policy that other labs are now navigating.

Shazeer's exit leaves Google with a bigger problem

For Google, the loss is more than a personnel story. Shazeer returned to Mountain View in 2024 specifically to lead Gemini, the lab's flagship effort to catch OpenAI on consumer and enterprise AI products. Two years later, the project is shipping faster than it was, but the model race has hardened into a three-front competition with OpenAI and Anthropic. Losing a co-lead at the moment the IPO clock is ticking for a primary rival is a hard signal to absorb internally.

The same week, Alphabet's stock closed lower after a separate report that several other Google AI researchers had taken meetings with OpenAI and Anthropic. It is the same pattern that played out earlier this year when Andrej Karpathy moved to Anthropic and became one of the most-watched lab moves of 2026. None of those departures have been confirmed at the same level as Shazeer's, but the pattern is now visible. Google has a structural problem in the talent market: the company's research prestige is high, its model progress is real, and its products reach billions of users, but for the small number of researchers who care more about advanced post-training work than distribution, the inside view of OpenAI's roadmap is a powerful draw. Shazeer's move makes that draw harder to resist for anyone sitting on the fence.

There is also a product question. Gemini's consumer rollout, including the Gemini app and the integration into Pixel and Workspace, has been one of the cleanest model-to-product pipelines in the industry. The next phase of Gemini work, on agentic capabilities and long-horizon tool use, is exactly the area where Shazeer's research taste is most valuable. Whoever replaces him as co-lead now inherits a project that is in a stronger public position than it was a year ago, but also one whose competitive window against OpenAI's agent roadmap has narrowed. The two hires OpenAI just disclosed are not two unrelated personnel stories. They are coordinated signals about how OpenAI wants to be perceived between now and the IPO. The first signal is technical credibility, with a co-author of the Transformer paper walking through the door two years after Google paid $2.7 billion to bring him back. The second is policy credibility, with a former White House AI architect reporting into the chief strategy officer on day one.

For anyone building on these models, the practical effect is small in the short term. Shazeer will not ship code this quarter, and Ball's Strategic Futures team is too new to publish anything substantive before the fall. But the practical effect in the medium term is large. OpenAI is assembling the team it wants to look like a public company, and the rest of the frontier is now responding. Watch for OpenAI's next few agentic infrastructure announcements to land in the same window as the S-1, and for Anthropic and Google to move more of their own senior researchers and policy staff into roles that read clearly to public market investors. For a deeper read on how the three frontier model families stack up for teams choosing a primary provider, our Claude vs GPT vs Gemini comparison walks through the same competitive picture from the buyer side.

According to reporting from Reuters on the same day, the move was confirmed by people familiar with the decision, and Google has not yet named a successor for Shazeer's Gemini co-lead role.