In partnership with
In Today’s Issue:
🩺 Midjourney builds a full-body ultrasound scanner
⚖️ The DOJ shields xAI's turbines as a national-security asset
🌐 Anthropic and DeepMind pitch a US-led AI coalition at the G7
🧬 OpenAI's LifeSciBench grades AI on real biology
🧠 An AI-built atlas of the body's aging "zombie cells"
✨ And more AI goodness…
⚡ The Signal
AI's frontier is quietly moving from the screen to the body. Today the AI story is mostly biological. Midjourney, of all companies, wants to put a radiation-free full-body scanner in a spa; OpenAI is grading models on real life-science research; and an NIH consortium used AI to map the "zombie cells" behind aging. Underneath all of it sits the plumbing that makes any of it possible: NVIDIA's optical supply chain, xAI's power-hungry turbines now defended as national-security infrastructure, and a G7 fight over who writes the rules. The throughline is convergence, with models reaching for health, science, and the physical world at the same time, which raises the stakes on both the claims and the oversight.
All the best,
Kim Isenberg
Mobile gas turbines at xAI's Colossus 2 data center near Memphis (Getty Images via WIRED)
⚖️ The DOJ Calls xAI's Turbines a National-Security Asset
The U.S. Justice Department asked a federal court to toss an NAACP lawsuit over xAI's gas turbines, arguing a shutdown would threaten national security. The case targets 46 unpermitted natural-gas turbines powering xAI's Colossus 2 data center in Southaven, Mississippi, just south of Memphis, which the NAACP (with the Southern Environmental Law Center and Earthjustice) says violate the Clean Air Act. The DOJ, now siding with xAI and the state of Mississippi, calls the site critical to the economy and the military, and a Department of Defense declaration says Grok's government model supports "vital national security missions," including recent U.S. strikes against Iran. Critics warn the move could become a template for AI firms to sidestep environmental permitting.
👉 tl;dr: Once a data center's turbines are branded a national-security asset, environmental law starts bending around AI infrastructure.
Anthropic CEO Dario Amodei at the G7 summit in Évian-les-Bains (New York Post)
🌐 Anthropic and DeepMind Pitch a US-Led AI Coalition at the G7
At a closed-door G7 lunch in Évian-les-Bains, Anthropic's Dario Amodei and Google DeepMind's Demis Hassabis urged democracies to build a US-led coalition to set international AI rules. The pitch landed days after the Trump administration imposed export controls on Anthropic's Fable 5 and Mythos 5 models on June 12 over national-security concerns, forcing the company to disable both worldwide. Amodei argued democratic countries should coordinate on frontier-model access, China-excluding semiconductor trade, and AI risks in cyber and bio; Canadian PM Mark Carney backed a US lead, while a European Commission spokesperson said the EU is "ready to engage." It also sketches the contours of a new Cold War: a US-led, China-excluding AI bloc on one side and Beijing's stack on the other, with compute and model weights as the dividing line instead of missiles.
👉 tl;dr: The G7 is quietly drawing the borders of a new Cold War, with AI chips and model weights as the front line.
NVIDIA CEO Jensen Huang at a 'Manufacturing for the AI Era' talk (NVIDIA)
🔌 NVIDIA's Optical Bet Breaks Ground in Texas
Coherent broke ground on an expanded fab in Sherman, Texas to mass-produce the indium-phosphide optics that move data between NVIDIA's AI chips. The site will house the world's first volume 6-inch indium phosphide (InP) line, wafers with roughly 4x the usable area of today's 3-inch ones, feeding the lasers and transceivers inside NVIDIA's Spectrum-X and Quantum-X Photonics switches with co-packaged optics, which in turn wire up systems like the 576-GPU Vera Rubin Ultra NVL576. It is backed by a $50M CHIPS Act grant and follows NVIDIA's $2B investment in Coherent. As Coherent CEO Jim Anderson put it, "AI scales on connectivity."
👉 tl;dr: The AI buildout is now a photonics race, with light, not copper, becoming the bottleneck between GPUs.
This issue is full of AI making big medical and scientific claims, a one-minute body scan, a life-science benchmark, an atlas of aging cells. Treat a frontier model as a translator and a skeptic, not an oracle.
Why it helps: Models are good at turning a dense paper, a benchmark table, or your own lab report into plain language, and at surfacing the caveats a headline skips, as long as you force them to cite sources and hedge.
Try this: "Here is a medical study, benchmark, or lab report: [paste]. Explain it in plain English in about 150 words. Then list (1) what it actually shows, (2) what it does not show, (3) the three biggest caveats or limitations, and (4) the questions I should ask a doctor or expert before acting on it. Flag anything that reads like marketing rather than evidence."
🎬 Watch This
NVIDIA just put its robotaxi brain in the open. Alpamayo 2 Super is a 32-billion-parameter reasoning vision-language-action (VLA) model for Level 4 robotaxis, unveiled at GTC Taipei. It triples the parameter count of the previous Alpamayo (from 10B), adds 360-degree perception and "Meta-Actions," and is built to reason through rare, long-tail driving situations instead of following hard-coded rules, which gives the interpretability regulators want for safety validation. NVIDIA says it ships this summer as open inference code on GitHub and downloadable weights on Hugging Face. The clip walks through how the model "thinks" before it drives.
"Resist the temptation to splinter."
– – Dario Amodei, CEO of Anthropic, urging G7 leaders to coordinate rather than fragment their AI rules (CNBC)
Meta's Superintelligence Labs is reportedly leaking talent. According to eWeek, the unit Mark Zuckerberg built with nine-figure pay packages has lost at least eight researchers, several heading straight to rivals: Avi Verma reportedly rejoined OpenAI after less than a month, and Rishabh Agarwal left for Periodic Labs after roughly five months. OpenAI looks like a frequent landing spot, which suggests that, so far, money alone has not bought stability inside MSL. Meta has not publicly detailed the departures.
Midjourney Wants to Scan Your Whole Body in 60 Seconds
The Takeaway
👉 Midjourney, the AI image company, is moving into hardware with a full-body ultrasound scanner and a new "Midjourney Medical" division.
👉 The pitch: a whole-body scan in about 60 seconds, with no radiation and no magnets, built on 40 Butterfly Network ultrasound-on-chip modules.
👉 The first product is non-diagnostic body-composition mapping (FDA general-wellness), not a replacement for MRI or CT.
👉 Rollout starts at a "Midjourney Spa" near San Francisco's Union Square by the end of 2027, with plans for 50,000 scanners and a billion scans a month.
Midjourney, the company best known for turning text prompts into images, just unveiled its first piece of hardware: a full-body medical scanner. On a June 17 livestream, founder David Holz introduced the Midjourney Scanner and a new Midjourney Medical division, calling it "the first new whole-body medical imaging modality in 50 years." You step into a water-filled chamber lined with submerged transducers, and the system builds a 3D picture of your insides from ultrasound, with none of the radiation of a CT scan or the heavy magnets of an MRI.
A labeled body-composition view from the Midjourney Scanner (Midjourney via R&D World)
The hardware is licensed from Butterfly Network, whose ultrasound-on-chip technology puts an entire probe on a CMOS semiconductor; each scanner packs 40 of those modules. The claims are bold: a full-body scan in about 60 seconds, at a prototype cost Midjourney says is roughly 10x cheaper and 60x faster than an MRI, which today runs $400 to $4,000 and takes one to two hours. The company frames it grandly: "We're building a bold new kind of machine to reimagine the foundations of healthcare and our relationships to our bodies."
The scanner's transducer ring imaging a cross-section of the body (Midjourney via R&D World)
The catch is what it actually does on day one. The first offering is body-composition mapping, a wellness feature (think fat, muscle, and organ outlines) that fits inside the FDA's general-wellness policy, not a diagnostic tool that flags tumors. Labels like "Ultrasonic CT" are marketing rather than new physics, and ultrasound still struggles to see through bone and air. Midjourney plans to open a "Midjourney Spa" near San Francisco's Union Square by the end of 2027, with ambitions to deploy 50,000 scanners and deliver a billion scans a month.
Why it matters: A consumer-AI company licensing imaging hardware to chase cheap, fast, radiation-free body scans says a lot about who builds medical devices now. The gap between a wellness body-composition map and a real clinical diagnosis is exactly where these claims deserve scrutiny.
The IT strategy every team needs for 2026
2026 will redefine IT as a strategic driver of global growth. Automation, AI-driven support, unified platforms, and zero-trust security are becoming standard, especially for distributed teams. This toolkit helps IT and HR leaders assess readiness, define goals, and build a scalable, audit-ready IT strategy for the year ahead. Learn what’s changing and how to prepare.
The chart: LifeSciBench is OpenAI's new benchmark of 750 expert-written tasks across real biology-research workflows, built with 173 scientists. The "Overall Scores" view puts OpenAI's science model GPT-Rosalind on top at roughly 63%, GPT-5.5 close behind near 59%, Google's Gemini 3.1 Pro around 52%, and Grok 4.3 last near 34%.
The lesson: Purpose-built scientific models still lead, but the field is bunched: the top three sit within about 10 points, and even the leader clears only about a third of tasks on the stricter pass-rate metric (GPT-Rosalind at 36.1%). "Useful lab assistant" is much closer than "automated scientist."
The caveat: It is OpenAI's own benchmark scoring OpenAI's own model, the headline "overall score" is a rubric aggregate that runs higher than the harsher task-pass-rate, and rankings flip by task (Gemini 3.1 Pro reportedly led on 214 of 750). Read it as a snapshot, not a verdict.
🧬 AI Maps the Body's Aging "Zombie Cells"
⚡ Bottom line: An NIH-backed consortium used AI to build the first cross-tissue atlas of senescent "zombie" cells, published June 11 in Cell.
💡 Why it matters: These non-dividing cells drive aging and disease, so mapping where they sit is the prerequisite for drugs that clear them.
🔎 What it means: Aging research shifts from "senescence is one state" to a tissue-by-tissue map, the data playbook that reshaped genomics.
The NIH Cellular Senescence Network (SenNet), launched in 2021, just released its first wave of results: the first comprehensive reference atlas of senescent cells across human tissues, mapped at single-cell and spatial resolution. The work is anchored by a cover commentary in Cell (June 11) led by Yale's Rong Fan with researchers from nine other institutions, alongside papers in Nature Aging, Nature Genetics, and Molecular Cell. Senescent cells stop dividing but stay metabolically active and pile up with age, earning the nickname "zombie cells" for the inflammation they leak into nearby tissue.
An atlas of senescent cell types mapped across human organs (Credit: Cell, 2026)
The headline finding is that senescence is not one thing. It is a spectrum of profiles the team calls "senotypes" that vary by tissue and disease, and AI and computational tools were used to classify aged-cell types and pull those senotypes out of the data. The consortium built new atlases for the brain (prefrontal cortex), liver, skin, lungs, and lymph nodes, and a Yale immunosenescence study found localized hotspots of dysfunctional B cells in older lymph nodes, a clue to why immune defense fades with age.
Yale's Rong Fan, who led the Cell cover commentary (Credit: Fan Lab, Yale)
The practical payoff is closer than a cure. The consortium also flagged blood markers that can help predict kidney disease, frailty, and future diabetes risk, pointing toward simple tests and, eventually, senolytics, drugs that selectively clear harmful senescent cells. As Rong Fan put it, "Cellular senescence is a fundamental hallmark of aging, yet we still know surprisingly little about where these cells reside in the human body." The atlas is a map, not a treatment, and the leap from cataloguing senotypes to safely clearing them in people is still ahead.
Lindy sorts your inbox, drafts replies in your voice, and preps you before every meeting over iMessage. 400,000+ professionals use it. Try it free.
