welcome to the first dispatch. once a week we read the AI news so you don't have to, and send back the three stories that actually moved, the reasons they matter, and one thing worth your weekend. no hype, no homework. this week the through-line wrote itself: talent, cost, and control all shifted at once, and not in Google's or Washington's favor.
1. Google lost its two most important researchers in 24 hours
On June 18, 2026 Noam Shazeer, a co-author of the 2017 Transformer paper ("Attention Is All You Need") and a Gemini co-lead, announced he was leaving Google for OpenAI. The next day, June 19, John Jumper, who built AlphaFold and shared the 2024 Nobel Prize in Chemistry, announced he was leaving Google DeepMind for Anthropic after nearly nine years. Two researchers of this caliber departing to two direct rivals within 24 hours is a striking signal about where insiders think the frontier is heading, right as Google's flagship is under competitive pressure. Markets reacted on June 22, with Alphabet shedding roughly $225 billion in market value (about a 5% drop; some outlets cited figures up to ~$270 billion).
2. An open Chinese model caught the frontier on coding, at a sixth of the price
Beijing's Z.ai (formerly Zhipu) released GLM-5.2 as MIT-licensed open weights, and it plugs straight into Claude Code as a credible daily driver. Nathan Lambert calls it open weights' DeepSeek-R1 moment and clocks it at roughly 6.8 months behind the closed frontier, squarely in the 6-9 month US-China gap. VentureBeat reports a one-million-token context window, a narrow edge over GPT-5.5 on SWE-bench Pro (62.1 to 58.6), and about a sixth of the API cost. An open model this good is real pricing pressure on the labs whose revenue depends on being the only option.
read the source at Interconnects (Nathan Lambert) →
3. The US pulled a frontier model offline, and now it's in court
A June 12 Commerce Department (Bureau of Industry and Security) directive forced Anthropic to disable Fable 5 and Mythos 5 for any foreign national, citing national-security concerns over a "narrow potential jailbreak" that could let the model identify software vulnerabilities. Anthropic shut the models off for all customers worldwide the same day to comply, and on June 23 legal-tech firm Legion LegalTech Corp sued in DC federal court to vacate the order, saying it cut off its Canada-based developers. The precedent is the story: Washington has now reached into a deployed commercial model and ordered it switched off for foreigners — a major escalation from export controls that historically targeted chips, not access to AI itself.
✶ one good thing
An AI model just disproved an 80-year-old math conjecture. Nature covered it and mathematicians checked the work: an OpenAI reasoning model generated the key ideas that disproved Erdős's unit-distance conjecture, a problem that had stood since the 1940s. This isn't retrieval dressed up as insight; it's a machine contributing an original step to a proof. Read it slowly. The questions it raises about who gets credit for an idea will outlast this week's news cycle.
that's the first one. thank you for being here at the start. see you next friday. — jordi, astral labs