Barret Zoph is back where he started after the Thinking Machines Lab co-founder announced on X that he had returned to Google DeepMind. Zoph’s return follows a six-year detour that included two stints at OpenAI and the launch of an AI lab with Mira Murati, who briefly replaced Sam Altman at OpenAI in 2024.

Zoph said he was excited to be “joining Google DeepMind!” as vice president of research. He also said he was optimistic about contributing to the lab’s reinforcement learning and post-training work, areas that have become central to how frontier labs refine and align models after pretraining.

The outside perception is less positive, however, with the AI lab appearing to be in the middle of a significant reshuffle, driven largely by departures at the top.

Where has Zoph worked?

Zoph first came into view when he joined the Google Brain Residency program. After about six years as a Google Brain research scientist, he moved to OpenAI, where he rose to the position of vice president of research.

DeepMind absorbed Google Brain in 2023, folding Google’s two principal AI research organizations into a single unit.

After a dramatic year at OpenAI, Zoph left in late 2024 to co-found Thinking Machines Lab with Mira Murati. He also served as chief technology officer there.

The Thinking Machines Lab chapter ended acrimoniously in January 2026 after allegations of misconduct surfaced involving a relationship with a colleague that was never disclosed.

That same month, Zoph returned to OpenAI alongside Luke Metz and Sam Schoenholz to lead the company’s enterprise push. That stint ended in June and lasted only five months.

Zoph’s research expertise lies in foundational areas such as neural architecture search and large-scale model training, rather than the product layer that has driven the AI talent wars. Even so, hands-on experience orchestrating large training runs remains scarce across the industry, which helps explain why researchers with his background move between frontier labs so frequently.

Google DeepMind is working to stay relevant in AI

Zoph arrives at Google DeepMind as the lab tries to refocus around its Gemini model, Google’s flagship AI family, which competes with rival systems such as OpenAI’s GPT and Anthropic’s Claude. That effort has been slowed by leadership setbacks in recent weeks.

Just three weeks ago, Demis Hassabis stepped back from his day-to-day DeepMind role to become chairman and Alphabet’s chief scientist in early August. CTO Koray Kavukcuoglu took over to fill the operational gap left by Hassabis.

Chief scientist Jeff Dean and senior staff members including Sanjay Ghemawat, Oriol Vinyals and Quoc Le left to start their own startup, Discovery Loop.

According to Fortune, Google’s corporate arm has also absorbed DeepMind teams.

Zoph’s arrival also comes against a difficult stretch for DeepMind. Data from UK intelligence firm Zeki, shared with Fortune, shows the lab’s ratio of research and engineering hires to departures fell from roughly 12-to-1 in the second quarter of 2023 to about 2-to-1 in the third quarter of 2026.

Among DeepMind staff who left in the past year, Zeki found that 25% went to Anthropic, 21% to Meta and 14% to OpenAI. The lab’s share of such hires across Europe, the Middle East and Africa declined from 49% in 2022 to 2023 to 18.6% in 2025 to 2026. Against that backdrop, Zoph’s return adds a senior research voice to a leadership group still taking shape under Kavukcuoglu.