Google DeepMind's rough stretch: four senior researchers gone in a week — three to Anthropic — as Gemini 3.5 Pro slips to July

What happened

Two things, in the same stretch.

The departures (per Bloomberg, Fortune, TechCrunch, Axios):

• John Jumper — AlphaFold lead and 2024 Nobel laureate in Chemistry — to Anthropic (announced ~June 19).
• Jonas Adler — worked on Google’s AI coding tools — to Anthropic.
• Alexander Pritzel — involved in pretraining — to Anthropic.
• Arthur Conmy — senior research engineer on Gemini 2.5 and AI safety — to Anthropic (posted June 25).
• Plus, a week earlier, Noam Shazeer — Gemini co-lead, Transformer co-inventor — to OpenAI.

Three of the four most recent exits went to Anthropic specifically.

The delay (per Business Insider, picked up by Reuters/Bloomberg): Gemini 3.5 Pro, the flagship Sundar Pichai promised “next month” at I/O on May 19, has reportedly slipped to July while Google refines it. Crucial caveat: Google has not officially confirmed this — a spokesperson declined to comment, and there is no announcement on blog.google. As of now, Gemini 3.5 Pro remains in limited Vertex AI preview; Gemini 3.5 Flash is the public 3.5 model.

Why this matters

Three reads.

1. Talent flows are a leading indicator — and the direction is unusually clear. Frontier AI has few honest signals; where elite researchers choose to work is one of them. Four senior DeepMind departures in a week, with three going to the same rival (Anthropic), isn’t random churn — it’s a pattern. It echoes Anthropic’s broader momentum (enterprise adoption, the Jumper hire signaling an AI-for-science push). When the people who built Gemini 2.5 and Google’s coding tools leave for Anthropic in the same week, the arrow points somewhere specific.

2. The delay and the departures compound each other. Either alone is survivable. Together, they write a narrative: a lab losing senior people and unable to ship its flagship on time. Perception matters in this market — it shapes recruiting (a virtuous or vicious cycle), enterprise confidence, and the framing investors apply as OpenAI and Anthropic head to public markets with cleaner momentum stories. Google also recently pushed consumers off Gemini CLI onto Antigravity, adding to a sense of churn.

3. But the contrarian case is strong, and worth stating plainly. DeepMind is one of the largest, deepest AI research organizations on earth, still led by Demis Hassabis. Gemini is the clear #2 chatbot with structural advantages no rival has — native Google Workspace integration, Search distribution, and its own TPU silicon. A delayed Pro launch driven by “we want it to be better” is not obviously bad; shipping a weak frontier model would be worse. Four departures from an organization of DeepMind’s size do not, by themselves, decide anything. The accurate frame is pressure and questions, not decline.

What it means for AI-tools buyers

If you use Gemini: nothing changes today, and you shouldn’t switch on the basis of org-chart news. Gemini remains excellent for Google-Workspace-centric work and research with citations — see the Gemini vs ChatGPT and Claude vs Gemini comparisons for where it wins. Judge it on what it ships.

If you were waiting for Gemini 3.5 Pro: plan for July, not June — but treat that as reported, not promised. If you need frontier-tier capability now, Claude Opus 4.8 and GPT-5.5 are available today; Gemini 3.5 Flash covers most needs at lower cost in the meantime.

If you’re making a longer-term platform bet: the talent and delivery signals are a reason to watch, not to abandon. Diversify rather than over-committing to any single lab’s roadmap — keep a primary and a fallback across Claude, ChatGPT, and Gemini. The whole point of our best AI chatbots guide is that the right pick depends on your work, not on this week’s narrative.

The honest caveats

The delay is not officially confirmed. This is the most important caveat. “July” comes from Business Insider/Reuters reporting citing sources, not from Google. A spokesperson declined to comment; blog.google has no post. It’s well-sourced reporting, but until Google says so or the calendar passes, treat it as reported, not fact.

Departures are real but DeepMind is vast. Four senior people is a genuine signal, but DeepMind employs thousands of researchers. Framing this as “DeepMind is collapsing” would badly overread it. Hassabis is still there; the bench is deep.

Talent moves are noisy. People leave for scope, autonomy, conviction, compensation, and timing. The Anthropic concentration is notable, but we don’t know each individual’s reasons, and inferring a single cause would be speculation.

Market-reaction figures are not clean. Some coverage tied Alphabet share moves to these stories; attributing specific market-cap swings to a handful of departures is unreliable, and we don’t anchor to those numbers.

What it changes for Pick Right readers

Practically, nothing about your tools today — Gemini works as well as it did last week. What changed is the story around Google’s frontier-AI program: a week of senior departures (three to Anthropic) and a flagship that’s reportedly slipping. The disciplined takeaway is to watch Google’s execution into July — does Gemini 3.5 Pro ship, and is it competitive? — while continuing to judge the products on merit. If you bet on AI platforms for the long run, this is a nudge toward diversification, not panic.

For the connected threads, see the Noam Shazeer → OpenAI move, the John Jumper → Anthropic move, the Gemini review, the Gemini vs ChatGPT comparison, the Claude vs Gemini comparison, the Google I/O 2026 recap (where Gemini 3.5 Pro was promised), and the best AI chatbots guide.

Questions or corrections? Email Pick Right. Want the full list? See all news.