Table of Contents
Welcome back to AI News Friday. 📰🤖
This was the week the AI world got real in ways nobody predicted. Washington didn’t just talk about regulating AI — it acted. A Friday export order forced Anthropic to shut off its most powerful models to users outside the US. OpenAI handed a cyber-defense version of GPT-5.5 to trusted defenders. And the biggest surprise? The AI image generation company announced it’s going into medical imaging.
Let’s get into it.
1. Washington Pulls the Plug on Anthropic’s Frontier Models
This is the biggest AI policy story of the year, and it happened on a Friday when everyone was trying to wrap up for the weekend. The Trump administration issued an export-control order that forced Anthropic to cut off all foreign access to Fable 5 and Mythos 5 — their two most advanced models.
The backstory is even more alarming. A US spy chief told the Senate that Mythos broke into almost all classified systems in hours — not weeks. The White House didn’t wait. They went straight to the nuclear option: block foreign access entirely.
Kenny’s Take: This is the precedent everyone’s been waiting for — and dreading. We’ve had chip export controls, model weight restrictions, and compute caps. But actually flipping the switch on a live, deployed frontier model for foreign users? That’s a new category of government power over AI. The technology community needs to understand: the window where frontier AI was a purely technical conversation is closed. It’s now a national security conversation, and the government has proven it will act — fast and decisively — when it feels threatened.
2. OpenAI Arms the Cyber Defenders
In a move that feels designed to contrast directly with Anthropic’s lockdown, OpenAI announced it’s giving the full GPT-5.5-Cyber model to “trusted defenders.” This isn’t a security-audited version with guardrails removed — this is a purpose-built cyber-defense model, handed to people who know how to use it.
The “Patch the Planet” initiative goes alongside it, a coordinated campaign to clean up critical infrastructure vulnerabilities using AI-assisted analysis. OpenAI is betting that open but gated access beats Anthropic’s everything-is-locked-down approach.
Kenny’s Take: Two AI labs, two completely different philosophies on security, and both happening in the same week. Anthropic gets its models shut off by the government; OpenAI proactively gives its cyber model to defenders before anyone can force the issue. OpenAI is playing chess here — they’re making themselves the responsible partner before the government decides for them. The “Patch the Planet” branding is also smart. Say what you will about Sam Altman, but the man knows how to frame a narrative.
3. Claude Tag Joins the Team in Slack
Anthropic also rolled out a smaller but interesting product update: Claude Tag for Slack. You can now @-mention Claude like a real teammate — it lives in channels, responds to threads, and has context about the conversations happening around it.
Separately, Mistral launched a cheaper document AI product, the White House issued a quantum computing executive order, and there’s a running debate about who actually leads in voice AI models.
Kenny’s Take: Claude in Slack is the kind of feature that sounds small but changes how people actually use AI day-to-day. The barrier to using an AI assistant at work has always been friction — opening a separate app, pasting context, waiting for a response. Putting Claude inside the conversation flow removes that friction entirely. It’s the same play Microsoft tried with Copilot in Teams, but Anthropic has the advantage of a model people actually want to talk to. Expect every AI lab to copy this within six months.
4. The Year AI Went Public
In a single quarter, SpaceX, Anthropic, and OpenAI all moved toward public markets. Combined, the private valuation pipeline behind these companies is worth roughly $3.7 trillion. To put that in perspective, the entire US IPO market raised only about $47 billion in the same period.
The public markets have never seen anything like this. These aren’t traditional tech IPOs — these are companies with completely novel risk profiles, regulatory uncertainty, and business models that barely existed five years ago.
Kenny’s Take: The $3.7 trillion versus $47 billion comparison tells you everything about the gap between private AI valuations and public market reality. Someone is going to get very rich, and someone is going to get very burned. The question is which companies are which. Anthropic going public after a week where the government literally shut off its models is… a choice. But if you’re an investor who believes frontier AI is the next electricity, the timeline doesn’t matter — you want in. The IPO market is about to get very interesting.
5. Midjourney Builds a Full-Body Scanner?!
This is the story I still can’t quite believe is real. Midjourney — the company that makes AI image generation tools — just unveiled a radiation-free, 60-second full-body ultrasound scanner and announced a new Midjourney Medical division.
Let me repeat that. The AI art company is now a medical device company.
Kenny’s Take: I keep coming back to the same thought: nobody saw this coming. Midjourney was the company making pretty pictures for designers and artists. Now they’re trying to compete with GE Healthcare. The through-line, if I squint, is computer vision and image processing — the same underlying technology that generates an image of a astronaut on a horse can theoretically reconstruct a 3D body scan from ultrasound data. But this is still a massive pivot. Either Midjourney has been quietly building a medical hardware team for years, or this is the most ambitious overreach since… well, since every other AI company tried to become everything at once. I’m genuinely curious which it is.
6. Did an ASML Chip Machine Reach China?
US Commerce Secretary Howard Lutnick told ASML that one of its extreme ultraviolet (EUV) lithography machines may have slipped into China. ASML insists that’s impossible — the machines are too large, too tracked, too regulated to end up in the wrong hands without anyone noticing.
This is escalating into a full standoff between the Commerce Department and the Dutch chip equipment maker.
Kenny’s Take: The ASML story connects to everything else this week. Export controls on chips, export controls on AI models — it’s all the same strategy. Control the inputs, control the outcomes. The fact that Lutnick is even suggesting an EUV machine reached China means one of three things: (1) they have intelligence ASML doesn’t know about, (2) they’re applying pressure to get ASML to tighten its compliance, or (3) someone smuggled the most tracked piece of industrial equipment on Earth into China without anyone noticing. Option 3 is the least likely but would be the most wild story of the decade.
7. Cognite’s Geir Engdahl on Agents in Production
Finally, a conversation worth reading. Cognite co-founder and CTO Geir Engdahl sat down for an interview about context, agents, and — most importantly — the gap between a pilot that works and a system people actually rely on.
It’s a refreshingly grounded take from someone who builds industrial AI for a living, where “vibe coding” won’t cut it.
Kenny’s Take: The industry loves talking about autonomous agents and AGI timelines. Engdahl is talking about the boring, hard work of making an AI system trustworthy enough that a refinery operator or oil rig manager will actually let it make decisions. That gap — between demo and deployment — is where 90% of AI companies die. Reading interviews like this is a good reality check when everyone is spiraling about Mythos breaking into classified systems or Midjourney sticking ultrasound wands on people.
⚡ Quick Hits
- Export controls on frontier AI — US forces Anthropic to disable Fable 5 and Mythos 5 for foreign users in a Friday executive order.
- OpenAI GPT-5.5-Cyber — Handed to trusted defenders alongside the “Patch the Planet” infrastructure cleanup initiative.
- Claude Tag in Slack — Anthropic turns Claude into a persistent Slack teammate you @-mention like a real colleague.
- Midjourney Medical — 60-second radiation-free full-body ultrasound scanner announced. Yes, seriously.
- AI goes public — SpaceX, Anthropic, and OpenAI all moving toward IPOs, backed by a ~$3.7 trillion private valuation pipeline.
- ASML standoff — US Commerce Secretary suggests an EUV chip machine may have reached China. ASML says impossible.
- Mistral cheaper document AI — French AI lab launches a more affordable document processing model.
- White House quantum order — New executive order on post-quantum cryptography standards.
Bottom line: This was the week that AI went from a technology story to a national security story — and nobody was fully prepared for the transition. The government proved it can and will shut down frontier models when it wants to. OpenAI tried to get ahead of the curve by arming defenders before the government forces their hand. And Midjourney turned around and did something so unexpected that it breaks the entire narrative about what an AI company is supposed to be. The next six months are going to be chaotic, unpredictable, and probably the most consequential period in AI’s short history as a public concern.
— Kenny