Fab’s Friday Field Notes — Week 17
Friday 25 April 2026
There’s a moment on every climb where the obvious hold is not the right one.
You’re three metres above your last piece of gear, your forearms are filling up, and there’s a big jug out to the right that looks like salvation. Everything in your body says grab it. But if you’ve read the route — if you’ve studied the sequence before committing — you know that jug takes you off-line. It pulls you into a position where the next move doesn’t exist. The right hold is smaller, worse, and directly above you. It doesn’t feel like progress. But it keeps you on the line that leads to the top.
I was thinking about this last week near Chiavenna, working a route that has exactly this problem at the crux. A good hold that leads nowhere, and a bad hold that leads everywhere. The climbing isn’t hard. The discipline is.
This week, Anthropic gave us a masterclass in the move you don’t make.
In case you missed it: in the space of ten days, they sent four signals that reshape the AI landscape. Their new model, Claude Mythos, became the first AI to complete a full 32-step corporate cyberattack simulation end-to-end — a task that takes human experts 20 hours. The UK’s AI Security Institute published the evaluation. The Bank of England is convening a CEO briefing. The Council on Foreign Relations called it an inflection point for global security.
Anthropic’s response? They didn’t release it. They restricted the model — a first in the industry — and built a defensive consortium (Project Glasswing) with Amazon, Apple, Google, JP Morgan, Microsoft, and Nvidia to use the capability to find and fix vulnerabilities rather than exploit them.
In the same week, they appointed the CEO of Novartis to their board through an independent governance trust that now holds the board majority. They acquired a 10-person biotech AI team for $400 million. And today, they announced an 800-person London expansion — choosing the UK as their primary European base, days after OpenAI did the same.
All of this while being blacklisted by the US Department of Defense for refusing to allow their technology to be used for autonomous weapons or mass domestic surveillance. A federal judge called the Pentagon’s retaliation “Orwellian.” The litigation continues.
I’ve written a longer piece unpacking all four signals — the cyber evaluation, the Narasimhan board appointment, the Coefficient Bio acquisition, and the London-vs-EU geography — with analysis at global, national, and personal level. You can read it here.
But for Friday, I want to stay with one thought.
Anthropic built the most capable AI model in history. The obvious move — the big jug out to the right — was to release it. Ship it to customers. Capture the market. Every incentive in the industry says: move fast, dominate, let someone else worry about the consequences.
They didn’t take that hold.
Instead, they held their red lines. They restricted the model. They built a consortium. They appointed a pharma CEO who has spent his career navigating the gap between breakthrough science and safe deployment. They chose London — a city that understands AI safety — over Washington, which was trying to punish them for having principles.
And in the same period, their revenue run-rate passed $30 billion. Their customer base doubled. Their valuation sits at $380 billion.
The move you don’t make is sometimes the move that defines you.
I think about this constantly in my own work. The partnerships I don’t take. The structures I build before the revenue arrives. The conflict-of-interest framework that exists before anyone has a conflict. The Red Lines document that says, clearly and in advance, what Fab Campaigns will and will not do — not because the opportunity isn’t there, but because the line is the line.
It’s the same discipline as the crux at Chiavenna. The obvious hold leads nowhere. The right hold feels worse. But it keeps you on the route.
Have a good weekend. Climb something. Hold your line.
This week’s Ideas piece: Four Signals From Anthropic That Should Change How You Think About Partnerships — full analysis of the AISI cyber evaluation, the Novartis CEO board appointment, the Coefficient Bio acquisition, and the London expansion, with implications at global, national, and personal levels.
Fab Campaigns Ltd · Partnership Architect · Winchester, UK