In partnership with
In Today’s Issue:
🤑 AI labs unite behind mandatory DNA synthesis screening
💵 Google brings multimodal Gemma 4 12B to laptops
📈 Medicine is moving faster across obesity, cancer and aging biology
📉 Anthropic’s IPO filing keeps Mythos risk in the spotlight
✨ And more AI goodness…
⚡ The Signal
The AI labs competing hardest on frontier models are now backing the same biosecurity rule.
OpenAI’s Sam Altman, Anthropic’s Dario Amodei, Google DeepMind’s Demis Hassabis, Microsoft AI’s Mustafa Suleyman and a long list of scientists, biotech executives and policy experts signed an open letter asking Congress to make nucleic acid synthesis screening and recordkeeping mandatory. The focus is the wet-lab supply chain, where AI-generated biological designs could become physical material. Model access is only one piece of that.
All the best,
Kim Isenberg
💻 Google Shrinks Multimodal Gemma to Laptop Size
Google introduced Gemma 4 12B, a mid-sized multimodal model designed to run locally with 16GB of VRAM or unified memory. The model uses an encoder-free architecture for vision and audio, supports Apache 2.0 licensing, and comes as Gemma models pass 150 million downloads. Running capable multimodal models locally, instead of only in the cloud, is becoming a practical option for developers.
👉 tl;dr: Google is pushing capable multimodal agents closer to everyday hardware instead of leaving them only in cloud inference.
🎨 Ideogram 4.0 Goes Open Weight
Ideogram released Ideogram 4.0 as an open-weight image model for design work. The model page emphasizes multilingual text rendering, precise layout control, editable elements, realistic 2K images and a training loop that teaches the system to read image structure before recreating it.
👉 tl;dr: The open-image-model race is moving from novelty generation toward design control, text accuracy and deployable weights.
🎓 Berkeley CS Classes Hit an AI Learning Wall
The Daily Californian reports that failing grades rose sharply in multiple UC Berkeley computer science classes in spring 2026. Instructors pointed to heavier AI reliance, weaker math preparation and understaffing as likely contributors, which is an uncomfortable signal for universities trying to separate AI-assisted output from actual mastery. The debate about whether AI will cause us to forget how to think is thus gaining new momentum.
👉 tl;dr: AI can make homework look better while leaving the underlying skill gap untouched until exams and projects expose it.
🧪 Biosecurity Stress Test
Ask your AI assistant to map any risky AI-and-biology story into four buckets: capability, bottleneck, control point and failure mode.
Try this: “Read this article or policy proposal and identify: 1) what new capability is being claimed, 2) what still prevents misuse in the physical world, 3) which chokepoint could be governed without blocking legitimate research, and 4) what evidence would change your risk assessment.”
🎬 Watch This
OpenAI’s seeded video resolves as “It’s time to fly | Codex.”
The teaser frames Codex as a broader building surface: “Welcome to the new frontier of building. Take on more tasks, move effortlessly across projects, and ship faster.”
The interesting bit is the positioning. Codex is being positioned as a workspace for moving through projects, tasks and shipping loops, a much bigger pitch than a simple coding helper.
“We also see early signs of recursive self-improvement (RSI) in today’s systems: where AI development is itself accelerated by AI. We expect this to increase competitive pressures among developers and nations, and create governance challenges that existing institutions are not equipped to address. As RSI emerges, societies will need ways to shape the trajectory of AI development and ensure that it serves human interests.”
– OpenAI, June 2, 2026, Democratic Governance of Frontier AI: A blueprint for a federal framework
Anthropic has confidentially filed for an IPO, according to TechCrunch, days after a $65 billion Series H reportedly pushed its valuation to $965 billion. The timing is awkward because Anthropic is also keeping Mythos restricted: TechCrunch says the model found thousands of high-severity bugs during testing and remains tied to limited cybersecurity access rather than a broad public release. That gives the IPO the feel of a market test for labs selling both explosive growth and unusually sensitive capabilities, a long way from a routine liquidity event.
AI Labs Ask Congress to Screen Synthetic DNA
The Takeaway
👉 OpenAI, Anthropic, Google DeepMind, Microsoft AI, Meta, Stripe, Y Combinator and major biosecurity names signed the same open letter.
👉 The letter asks US legislators to make nucleic acid synthesis screening and recordkeeping mandatory.
👉 The chokepoint is practical: synthetic DNA/RNA orders, synthesis equipment, customer verification and records that can support biosecurity investigations.
👉 The policy pitch is narrow enough to unite rival AI labs because it targets the physical supply chain, not a blanket restriction on AI research.
The strange thing about this letter is who agreed to sign it. Sam Altman, Dario Amodei, Demis Hassabis and Mustafa Suleyman spend most weeks competing for models, talent, enterprise customers and policy influence. Here they are on the same side: asking Congress to require screening for synthetic nucleic acids and the equipment used to make them.
The mechanism is specific. Screening would require providers of synthesized DNA and manufacturers of synthesis machines to check orders for sequences of concern and verify customer legitimacy before shipping. Recordkeeping would preserve synthesis orders and sequence data so that suspicious activity can be traced later, even when a single sequence does not look dangerous by itself.
The AI angle is what makes the old biosecurity problem newly urgent. ScreenDNA’s letter says AI systems now outperform PhD-level virologists on some highly technical lab-procedure questions, while Wired reports that signatories are worried AI could lower the knowledge barrier for designing toxins or pathogens. The letter does not claim every dangerous design can be made by typing into a chatbot. It argues that the supply chain needs a mandatory backstop before model capability and cheap synthesis move further ahead of voluntary safeguards.
Why it matters: The most credible AI safety fights are moving toward concrete bottlenecks: model evaluations, cyber access, DNA synthesis, customer screening and audit trails. Rival labs rarely agree in public unless the risk is obvious enough, or the fix is narrow enough, that nobody wants to be the holdout.
Sources:
🔗 https://gizmodo.com/sam-altman-and-dario-amodei-agree-for-once-sign-letter-against-ai-assisted-bioweapons-2000767292
🔗 https://screendna.org/
🔗 https://www.wired.com/story/openai-anthropic-letter-ai-biological-weapons/
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: Did Mustafa just leak the Claude Mythos FLOP count? On the chart shown at Microsoft Build keynote it’s on par with Gemini 3.1 Pro.
The lesson: If the plotted comparison is accurate, the frontier gap is no longer just about who has the biggest public model card. Private capability, restricted cyber access and selective disclosure are becoming part of the model-release story.
The caveat: Treat this as a leak-shaped signal, not confirmed benchmark evidence. The chart is useful because it captures the rumor; it should not be read as an independently verified Anthropic disclosure.
Medicine Is Moving Faster Than the Old Timelines
⚡ Bottom line New obesity drugs, personalized cancer vaccines and GLP-1 aging research are all pointing in the same direction: medicine is starting to compress timelines that used to feel fixed.
💡 Why it matters AI is one layer in a larger shift: computational biology, better trial infrastructure, richer biomarkers and faster drug-development loops are making progress feel less incremental.
🔎 What it means Near term, the evidence is about measurable movement on diseases that used to define whole decades of public health: obesity, melanoma recurrence, pancreatic cancer, inflammation and biological aging markers.
The clearest hard number comes from Lilly. In TRIUMPH-1, participants taking 12 mg retatrutide lost an average of 70.3 lbs, or 28.3% of body weight, over 80 weeks. Among participants with baseline BMI ≥35 who continued into the extension, the highest-dose group reached an average loss of 85.0 lbs, or 30.3%, at 104 weeks. Lilly is still describing retatrutide as investigational, but the result is already in the range people used to associate with bariatric surgery.
The cancer story is moving, too. NBC reports new five-year follow-up data for personalized melanoma vaccines, and the wider oncology coverage points to the same shift: cancer treatment is getting more individualized, more immune-system-aware and faster to validate once a signal appears. The stack around the medicine now matters almost as much as the medicine itself: sequencing, biomarkers, trial design and model-assisted discovery all shorten the path from hypothesis to patient data.
GLP-1s are also escaping the narrow frame of weight loss. A UC San Diego-linked study reported randomized, placebo-controlled evidence that semaglutide slowed the accumulation of biological aging markers in DNA among adults with HIV-associated lipohypertrophy over a 32-week treatment period. GLP-1s aren't anti-aging magic. What has changed is that metabolic drugs are now being studied through inflammation, epigenetic clocks, cardiovascular risk, organ health and healthspan.
The pattern is hard to ignore: medicine is becoming more programmable. Drugs, vaccines, diagnostics and trial readouts are starting to move through tighter feedback loops. The caution is equally important: every one of these signals still has to survive larger trials, longer follow-up, safety data and real-world delivery.
Sources:
🔗 https://x.com/erictopol/status/2061429974566093247?s=46
🔗 https://www.nbcnews.com/health/cancer/personalized-vaccine-melanoma-cut-risk-cancer-returning-five-years-rcna347424?cid=sm_npd_nn_tw_ma&taid=6a1e9b9b3df0d800015bf9d4&utm_campaign=trueanthem&utm_medium=social&utm_source=twitter
🔗 https://investor.lilly.com/news-releases/news-release-details/lillys-triple-agonist-retatrutide-delivered-powerful-weight-loss
🔗 https://www.healthline.com/health-news/eli-lilly-new-drug-retatrutide-shows-greater-weight-loss-results
🔗 https://x.com/GIMedOnc/status/2061078272465666504
🔗 https://x.com/cremieuxrecueil/status/2060955532559581294
🔗 https://x.com/DKThomp/status/2061110056293106118
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.