AlphaFold creator and Nobel laureate John Jumper leaves Google DeepMind for Anthropic: One week after Transformer co-author Noam Shazeer joined OpenAI, Google has lost another defining AI researcher. Jumper, who shared the Nobel Prize in Chemistry with Demis Hassabis for AlphaFold, spent nine years at DeepMind. Two symbolic Google AI researchers departing to competitors in the same week marks a new phase in the talent war — this is no longer about hiring engineers, it’s about who owns the scientists. Anthropic, gearing up for its IPO, is landing consecutive blows against both Google and OpenAI. TechCrunch · Reuters
Trump reverses on Anthropic after G7: “not a national security threat anymore”: In an Axios interview, Trump stated Anthropic has “behaved very responsibly” and signalled he may ease restrictions on its Mythos and Fable models. This marks a sharp reversal from three months of aggressive administration pressure. It reflects Anthropic’s diplomatic success at the G7 — and a reminder that AI export controls are functioning as political bargaining chips, not purely technical decisions. The Next Web
Microsoft reveals ‘AutoJack’: a single webpage can achieve RCE on AI agent hosts: Microsoft’s security team demonstrated a vulnerability where attackers can gain remote code execution on machines running AI agents — just by having the agent visit a malicious webpage. As AI agents move rapidly into production, agent host security is a vastly underappreciated attack surface. Every team deploying agents should take note. Microsoft Security
‘In the Weights’ launches — an AI-centric vanity search for the model era: Built by a former Instagram engineer, this tool lets you search what major AI models “know” and “think” about you or your company. Think of it as Googling yourself, but for the latent space of frontier models. A clever product that captures a growing anxiety: what do the models actually have on us? TechCrunch
Big Tech’s AI capex hits $750B combined in 2026, up 80% YoY — and the bond market is paying attention: Amazon, Alphabet, Microsoft, and Meta are projected to deploy a combined $750 billion on AI infrastructure this year. Companies are depleting cash reserves and leveraging debt. As one analyst put it: “Tech investors are learning what it’s like to be investors in old-economy, capital-intensive industrial businesses.” The AI infrastructure race is reshaping capital markets themselves. CNBC
AI agent security auditing is an emerging niche: AutoJack demonstrates that AI agent hosts represent a new attack surface not covered by conventional web security tooling. As agent deployments grow, there is a clear gap for B2B security audit tools and best-practice frameworks purpose-built for agent infrastructure. Microsoft Security
The second wave of the AI talent war is about scientists, not just engineers: After Shazeer (engineering) and now Jumper (fundamental science), the pattern is clear: the next AI frontier requires deep domain expertise married to model capability. Early-stage opportunities exist in building domain-specialized AI research teams — in biology, materials science, chemistry — where the science talent is the real moat. TechCrunch
The Anthropic export control whiplash: Trump’s verbal reversal is not yet policy. Meanwhile, JPMorgan and Goldman Sachs have already preemptively restricted Anthropic model access for Hong Kong staff. The gap between political rhetoric and market reality will be the key variable heading into the new week. The Next Web