DeepMind loses elite talent to rivals as Google internally tests Gemini 3.8 Flash

A Fortune analysis published August 27 reports that Google DeepMind is losing elite AI talent to rival labs and startups, citing Gemini co-lead Noam Shazeer's move to OpenAI and AlphaFold creator and Nobel laureate John Jumper's departure to Anthropic, among others. The exodus fuels a narrative that research-heavy staff are gravitating toward rivals even as DeepMind ships aggressively.

On the product side, Business Insider reports Google employees are internally testing a 'Gemini 3.8 Flash Preview,' continuing an unusually fast Flash cadence — Gemini 3.7 Flash arrived roughly three weeks after 3.6 with 1M-token context and aggressive introductory pricing ($0.75/$3.75 per 1M tokens). Separately, Google Cloud launched Gemini Enterprise for Legal, co-developed with law firm Weil, to automate legal workflows using AI agents — a vertical-specific enterprise push mirroring Anthropic's Claudeforce.

The juxtaposition is the point: Google is simultaneously the fastest shipper in the frontier race and the biggest talent donor to its competitors. That tension raises questions about whether Gemini's velocity is sustainable if senior researchers keep leaving, and whether the Flash line's rapid iteration reflects strength or contributes to the 'release velocity fatigue' enterprises complain about (roughly 11 days between frontier releases industry-wide).

Competitively, Gemini Flash remains a value leader that developers praise — 'at $0.75 per million input tokens it beats Claude Sonnet 5 and GPT-5.6 Terra on production code' — though introductory pricing reportedly expires December 31, 2026. Watch whether Gemini 3.8 Flash launches publicly, how Google counters the talent narrative (retention packages, new hires), and whether Gemini Enterprise for Legal wins reference customers against Microsoft and Anthropic in regulated verticals.