Open Science, Closed Doors, and Infrastructure Bets

Good morning. This week's stories sketch a familiar pattern: breakthroughs shared freely, security boundaries tested, and capital flowing toward the layers that make AI work at scale.

Google DeepMind released AlphaGenome into the commons, handing geneticists a tool that took years to build. Crunchbase learned that voice phishing still defeats strong authentication when the human answers the phone. Nvidia is staking two billion dollars on the data centers that train the models, and Synthesia doubled its valuation by turning those models into corporate training videos.

No single throughline ties these four together, but they do show the industry moving on three parallel tracks: research opening up, perimeters staying porous, and infrastructure attracting the serious money. The stories are below.