In partnership with

Welcome, AI enthusiasts
The AI model Dario Amodei fears most is one nobody can download. After a week of rivals accusing Anthropic of wanting Chinese open models banned to protect its own business, and his reply was that a ban would protect him and he still does not want one. What he asked for in its place reaches further than any download ban would, and it lands on anyone who runs a model on their own machine. Let's dive in!
In today’s insights:
Anthropic's CEO Fears an AI Made Pandemic
Nvidia Buys Into an OpenAI Co-Founder's Guarded Research
OpenAI's Ex-Safety Chief Quits Over Her Health
Read time: 4 minutes
LATEST DEVELOPMENTS
OPEN MODELS
🦠 Anthropic's CEO Fears an AI Made Pandemic

Source: Economic Times
Evolving AI: Dario Amodei set a bioweapon fear at the center of Anthropic's answer on open AI models.
Key Points:
The post calls open models without dangerous capabilities a public good, since they cost nothing beyond the compute to run them.
Amodei names two fears, authoritarian states building more powerful models than the US, and models capable enough to run cyberattacks or weaponize a pandemic-level virus.
He argues biology may favor the attacker, since the materials are widely available and any defense would take years.
Details:
Nvidia's letter Open Weights and American AI Leadership drew a flat denial from Amodei's post, that Anthropic has never advocated for a ban. What he does want is chip sales to China blocked and a legal crackdown on industrial-scale distillation, which he says can otherwise pull Chinese models to within a few months of the US frontier. Safety testing before release would then apply to every sufficiently capable model, whether it is open or closed.
Why It Matters:
Meta and Google both publish open weights, and everything already downloaded stays downloadable, since Amodei's own argument is that a released model cannot be pulled back. His testing bar would land on what comes next, so the strongest free releases would arrive later or not at all, and whoever runs those tests ends up deciding what anyone can put on their own machine.
TOGETHER WITH DATADOG
🔐 AI Security Best Practices
Evolving AI: Your Guide to Building Secure AI Applications.
As AI adoption accelerates, new attack surfaces are emerging across infrastructure, supply chains, and model interfaces. Datadog’s AI Security Best Practices Guide breaks down how to secure:
The underlying components that host and run AI applications
The software and data that an AI application uses to operate
The entry points and business logic that enable a user to interact with an AI application
This guide provides actionable strategies to help teams strengthen AI security without slowing innovation.
AI RESEARCH
🔓 Nvidia Buys Into an OpenAI Co-Founder's Guarded Research
Evolving AI: Ilya Sutskever gets ten times the computing power within twelve months of letting Nvidia in.
Key Points:
Safe Superintelligence has taken an undisclosed investment from Nvidia under a long-term partnership.
That power arrives through Vera Rubin, Nvidia's next-generation platform.
SSI has released no product since it launched in 2024, yet carries a $32 billion valuation.
Details:
Jensen Huang tied the decision to Sutskever's record, pointing to AlexNet as the foundation of modern AI. Nvidia says it committed only after rare access into work the lab has never shown outsiders, and Bloomberg reported the check near $5 billion. Sutskever's only comment was that the research is worthy of scaling up. The agreement also runs the other way, with SSI's findings shaping how Nvidia advances its current and future compute platforms.
Why It Matters:
Nvidia’s Vera Rubin will now be shaped in part by a lab whose entire purpose is keeping powerful AI aligned, and that shaping reaches the chips under the assistants people open every morning. Whatever safety thinking comes out of that lab arrives in their hands as hardware they never picked.
The browser that reads the room before you ask.
Most browsers get you to the page. Norton Neo gets you to the answer. Magic Box understands your intent before you finish typing — no prompting, no switching apps, no copy-pasting. Built-in AI, instantly and for free. Privacy handled by Norton, by default.
AI TALENT
🩺 OpenAI's Ex-Safety Chief Quits Over Her Health
Evolving AI: Lilian Weng is leaving the lab she cofounded with Mira Murati, days after its first model shipped.
Key Points:
At OpenAI she built the Safety Systems team from scratch after GPT-4 and grew it past eighty people.
Seven months of sickness, the worst stretch of her life, made the pace of a frontier startup impossible to hold.
AI research is still what she wants, just in a smaller role at a calmer company..
Details:
Weng's note describes taking sick leave while the team sprinted toward the launch of Inkling, and resting through it while carrying the guilt. She weighed shrinking her responsibilities before deciding that was not who she was, writing that failing to give the job her full effort means failing at it. Fidji Simo left her full-time role as OpenAI's second in command nineteen days earlier, after three months of medical leave.
Why It Matters:
Simo's departure and this one both came down to health, and the models kept shipping on the same schedule either way. What a chatbot will refuse to do for someone at two in the morning was set by a small number of people who understood the stakes, and Weng spent years as one of them.
QUICK HITS
🇨🇳 China's PsiBot Hits $1.48B Valuation in Latest AI Robotics Round.
💰 Nvidia in Talks to Guarantee ~$250B for OpenAI's Ohio Data Center Lease.
🛡️ AegisAI Raises $36M Series A for AI-Native Cybersecurity.
👓 Apple Delays Smart Glasses to WWDC 2027 Over Privacy Concerns After Meta Backlash.
📈 Trending AI Tools
🎞️ Guideless - AI tool that automatically turns your clicks and workflows into step-by-step video guides*.
✍️ Clico - AI writing assistant for every browser text box.
✂️ Klap - Turns long YouTube videos into viral TikToks, Reels, and Shorts.
🎥 Pictory - Turns scripts, blogs, and long videos into short branded clips.
*partner link






