Anthropic Got Hit by Export Rules Nobody Understands
The full picture of Anthropic’s week from hell is coming into focus: the Trump administration abruptly ordered Anthropic to cut off all foreign nationals — including its own employees — from Fable 5 and Mythos 5, forcing the company to take its newest models completely offline. The trigger, per Wired, was SK Telecom’s alleged ties to China, which apparently was enough to nuke global access for everyone. This is what happens when export control frameworks written for semiconductors get applied to software APIs — you get a blunt instrument wielded with all the surgical precision of a sledgehammer.
The White House Wants Anthropic to Block All Jailbreaks. That May Not Be Possible.
If Anthropic wants to rerelease Fable 5, the Trump administration is apparently demanding the company guarantee its models can’t be jailbroken — a requirement that every serious AI security researcher says is technically impossible. Asking a company to certify that no one, ever, can circumvent an LLM’s guardrails is a bit like asking a car manufacturer to guarantee no one will ever speed. The practical effect here may be less about security and more about giving the administration a permanent veto over Anthropic’s model releases.
World Leaders Want American AI. They Just Don’t Want America to Be Able to Turn It Off.
Macron and Modi raised alarms at the G7 that the U.S. could cut off access to American AI overnight — and then, almost on cue, Anthropic’s blackout made that fear completely real. This is the geopolitical consequence nobody was taking seriously enough: foreign governments and businesses building on American AI infrastructure are now operating with a kill switch they don’t control. The Anthropic situation just handed every open-source AI advocate and every Chinese AI company the best recruiting pitch they’ve ever had.
Midjourney Medical Goes from Generating ‘Cat Images’ to Full-Body Ultrasound Scans
CEO David Holz unveiled Midjourney’s first-ever hardware product: a ring-of-sensors ultrasound-based full-body scanner, plus plans for a San Francisco spa to house it. To his credit, Holz himself acknowledged this is a bit of a pivot from the company’s AI image generator roots. I genuinely have no idea what Midjourney’s long-term thesis is anymore, but I respect the audacity of a company that goes from “diffusion model for generating fantasy landscapes” to “we’re opening a wellness spa with diagnostic hardware.”
Google’s Medical AI AMIE Matches Primary Care Physicians in Complex Disease Management
New research published in Nature shows Google’s AMIE conversational AI system performing at parity with primary care physicians in managing complex disease cases. Getting a result like this into Nature is not nothing — this is peer-reviewed, not a press release benchmark — but “matches physicians in a research setting” and “ready for your doctor’s office” are still very different sentences. Worth watching closely, especially alongside Ontario auditors finding that 60% of AI medical scribe systems mix up prescribed drugs.
OpenAI’s Near-Autonomous AI Chemist Improves a Challenging Medicinal Chemistry Reaction
OpenAI and Molecule.one have demonstrated a near-autonomous AI chemist using GPT-5.4 that actually improved a key drug-making reaction — not just suggested improvements, but executed them in the loop. This is the kind of narrow, high-stakes domain application where AI’s ability to iterate rapidly through hypothesis space could genuinely compress drug development timelines. Less flashy than a chatbot demo, more likely to matter in ten years.
Two-Thirds of Americans Think AI Is Advancing Too Quickly
Pew Research finds 49% of Americans now use AI chatbots at least occasionally — up from 33% in 2024 — but 63% think the technology is advancing too fast. The gap between adoption and comfort is the most interesting data point here: people are using this stuff even though it makes them uneasy, which is less an endorsement of AI than a description of how technological inevitability actually feels from the inside.
Claude Code Costs Up to $200/Month. Goose Does the Same Thing for Free.
Block’s open-source Goose coding agent is mounting a real challenge to Claude Code’s pricing model, which tops out at $200/month — a number that’s generating genuine sticker shock among the developer community. The AI coding assistant market is about to get a brutal commoditization reality check: when the underlying models are accessible via API, the moat around a terminal-based wrapper is thinner than it looks. Anthropic has a brand and tight integration working for it, but “free and open-source” is a powerful counterargument.
Bottom Line
The Anthropic export control saga just became a masterclass in how government policy can simultaneously be technically incoherent, geopolitically destabilizing, and completely unenforceable — all at once.