You might have missed this last week, because the news cycle moved on fast. But something happened on Friday June 13th that I think we'll all be referring back to for years.

The US government ordered Anthropic to suspend access to its two most advanced AI models - Fable 5 and Mythos 5 - for any foreign national, whether they're sitting in Berlin, Lagos, Mumbai, or quite literally working inside Anthropic's San Francisco office on a non-US passport.

Because Anthropic has no way of verifying citizenship at sign-up, the only way to actually comply was to disable both models for everyone, globally. Even paying enterprise customers. Even Anthropic's own staff. The most capable AI in the world, available to the public only days ago, switched off in a single evening.

Most coverage treated this as a technology story. It isn't, really. It's the moment a quiet shift that's been building for two years became impossible to ignore - and if you're a founder, an investor, or anyone building something that touches AI, it changes a few assumptions you probably haven't questioned in a while.

Let me walk through what actually happened, what it means, and why I think this is a turning point we'll all be referring back to.

See below to continue…

Join the waitlist for SoPhy - the networking intelligence engine!

Soon, you'll be able to quantify your business relationships - but for now, we're starting with the essentials:

Quick update: the app is almost ready - just a few finishing touches, and we should be able to open it up in the next 2 weeks. Here is a short demo video to check out ahead of the release.

Join the waitlist to be one of the first to try it once it's out.

Sign up here.

🚀 Healthcare Accelerator from OVHcloud

Applications are open for our 10-week, equity-free program built for Healthcare founders ready to scale fast and connect globally.

🗓 Key dates

💡 What's included:

Whether you're refining your architecture, sharpening your GTM, or seeking investment - this is your launchpad.

👉 Apply now

Continue…

Anthropic released Fable 5 earlier this month, claiming it represents a new level of capability it calls "Mythos-class". Fable is the public-facing model with guardrails; Mythos is the underlying model the guardrails are wrapped around.

The model was particularly effective at identifying software vulnerabilities - which is genuinely useful for cybersecurity defenders and genuinely concerning if it falls into the wrong hands.

A few days after release, an administration official told Axios the Commerce Department decided to take action after another company claimed it was able to jailbreak Mythos, alarming the administration about possible national security risks. The administration had reportedly tried to get Anthropic to pause releasing the latest models but was unsuccessful, prompting the export control letter.

Anthropic, for its part, has pushed back hard. They've said the jailbreak is "narrow" and "non-universal" and that the same capability is already widely available in other models including OpenAI's GPT-5.5, which is not subject to any equivalent restriction. They argue that if this standard were applied across the industry, it would essentially halt all new model deployments for every frontier model provider.

They have a point. But it's a point that matters less than the precedent.

Here's the part I keep coming back to.

For two years, Anthropic has been the AI lab that loudest insisted its own technology might be civilisation-ending. Every press release has framed Mythos and its predecessors as something close to a weapons-grade capability that only Anthropic could be trusted to release responsibly.

And then the government took them at their word.

As a cybersecurity researcher put it on X this week:

That sentence will follow Anthropic around for a long time, because it's true. You can't simultaneously market your model as world-changingly dangerous and be surprised when the world's largest national-security apparatus treats it that way. The same safety-first positioning that earned Anthropic premium pricing, premium investors, and a $183B-plus valuation is now the exact reason the company has a Commerce Department export licence wrapped around its product roadmap.

There's a wider lesson buried in this for any founder who has ever wondered whether to lean hard on the "this is dangerous" angle in a pitch. The narrative you build about your own product is also the narrative regulators use to write the rules. Choose carefully.

The Commerce Department's letter is being framed as a one-off, but it's not. It's the formal start of something most of us have been quietly assuming was a few years away.

Frontier AI is now being treated as a national security asset by the world's largest government.

as Axios put it. And once intelligence has borders, a lot of things downstream of that change.

If you're building a product on top of US frontier models from anywhere outside the US, your access can be revoked overnight by a government you don't vote for, based on a security review you never see. That's now a real, demonstrated risk, not a theoretical one.

If you're a US company selling globally, you can no longer credibly promise enterprise customers in Europe, the Middle East, or Asia that your AI capabilities will be uniformly available. You can promise it on Monday and have it taken away on Friday.

And if you're an allied government, even one as friendly as the UK, you've just been shown - in public, in a single news cycle - that the most capable AI in the world is now a US-only resource, granted and removed at Washington's discretion.

That last point is the one that matters most.

If you're sitting in Paris, London, Berlin, Tel Aviv, Bangalore, or Dubai right now and trying to decide how much of your national strategy should rely on US frontier models, you just got your answer.

The only rational response to what happened this week is to assume that any US frontier model can be export-controlled at any time, on terms you don't influence and timelines you can't predict. Which means every serious country and every serious sovereign fund now has a strong motivation to do one of two things: build their own frontier models, or lean harder on Chinese open source models that no one in Washington can switch off.

Mistral in France was already moving in this direction. The UK's Sovereign AI Fund (which I wrote about a few weeks ago) makes more sense today than it did last Monday. India and the UAE have both been building national AI capacity quietly. Every one of these efforts just had their political case made for them, for free, by the US Commerce Department.

The likely outcome is the one nobody at OpenAI or Anthropic wants. A more fragmented global AI ecosystem. Three or four sovereign-flavoured frontier model stacks. And a meaningful share of the world quietly defaulting to whatever open source Chinese models look "good enough" - because at least they can't be turned off by a foreign government.

Three things to take from this week, regardless of where you sit.

For founders, single-vendor AI dependency just became a board-level risk. If your entire product stack runs on one US frontier model, you no longer have a continuity plan - you have a single point of geopolitical failure. Diversifying across providers, keeping an open source fallback warm, and architecturally designing for model substitution should be on the roadmap by Monday morning.

For investors, the IPO narratives for Anthropic and OpenAI just got more complicated. Anthropic is widely expected to pursue an IPO this year and has staked much of its public identity on being the safety-conscious alternative to its rivals. A model maker that can be effectively switched off by a single Commerce Department letter is, by definition, not a clean trillion-dollar global story. The valuation conversation will have to grapple with this. Quietly, but it will.

For everyone else, the most useful frame is this: intelligence is no longer a neutral utility we all assumed we'd get to share. It's an asset, like oil or semiconductors, and it now has the geopolitics to match. The countries, companies, and founders who understood this earliest are about to look very prescient.

A few months ago I wrote about how the frontier labs were quietly eating the application layer. A few weeks ago I wrote about how European VC was being reshaped by AI mega-rounds. This week the story took its next predictable step: governments are now reshaping the frontier labs themselves.

The arc is clear if you follow it. AI is no longer just a market. It's a geopolitical instrument, regulated like one, contested like one, and increasingly distributed like one.

That last word matters. Because what this week really showed us is that intelligence will not be distributed equally. Some people will get the best models. Others will get the leftovers. And the line between the two groups will be drawn by people who don't know you and don't ask your permission.

If you're not already diversifying your AI strategy, your stack, and your assumptions - today is the day to start.

The Anthropic story dominated my feed for two days. But there were three other things that happened in the same news cycle that are worth understanding, because each one tells you something about where this whole industry is going.

SpaceX listed on Nasdaq on June 12th under ticker SPCX. The shares priced at $135, opened at over $150, and closed the first day at $161.11 - a 19% gain that pushed the company's market cap to roughly $2.1 trillion and made Elon Musk the world's first trillionaire. Only about 4.3% of the company was floated, which is the strange bit.

Wall Street effectively priced the entire restaurant after Musk sold a single slice of pizza. Whether the valuation holds depends almost entirely on whether you believe the Starlink + space data centre + AI compute story, which is what justifies most of the premium above the rocket business itself. It also matters because SpaceX is now sitting at the intersection of telecoms, AI infrastructure, and defence in a way no other public company is - and the IPO sets the comp for the Anthropic and OpenAI listings expected later this year.

The Regional Court of Munich ruled that Google is directly liable when AI Overviews makes false statements - in this case, after the AI confidently linked two Munich publishers to scams, subscription traps and shady business practices, none of which appeared in any of the cited sources.

The court's logic is the bit that matters: AI Overviews isn't a search result, it's Google's own statement, "an independent presentation for which Google is responsible."

Google is appealing, but if the ruling holds, every AI Overview that says something false about a real person or business becomes a potential liability. Multiply that across the EU's population and Google's search volume and you can see why this could force Google to either disable AI Overviews across Europe or completely rebuild how the product cites sources. Hallucinations have just stopped being a quirky product flaw and become a balance-sheet item.

Mark Zuckerberg quietly reassigned roughly 6,500 Meta engineers and product managers into a three-month-old "Applied AI" unit, mostly via surprise email, with no real option but to join or quit.

Their daily work? Generating handwritten puzzles and coding problems to train Meta's AI models - because synthetic data has hit a quality ceiling and the only way past it is human-generated training material.

One employee told Wired the work was "literally the gulag." Another called it "soul-crushing."

Over 1,600 Meta employees company-wide have also signed a separate petition against the company monitoring their clicks and keystrokes for AI training data. Zuckerberg has since admitted Meta "made mistakes" in the restructure. T

he deeper signal is the one nobody is naming clearly: even at Meta's scale, training the next generation of models has hit a wall that money alone can't get past. And the smart, expensive engineers who built the platform are now being conscripted into generating the data that's being used to train their replacements. The irony writes itself.

Three very different stories. One common thread. Across all of them, AI is no longer a clean upside narrative for the people building it. It's geopolitically contested, legally exposed, and internally combustible at the company level. The trillion-dollar story is real. So are the trillion-dollar cracks.

✅ Know a founder or investor who hasn't fully grasped what happened this week? Forward this their way. The earlier they understand that intelligence now has borders - and that the cracks underneath the trillion-dollar AI story are real - the better the decisions they'll make over the next 12 months.

(👉 Vote now — we’ll share the results in next week’s issue. All votes are anonymous.)

(Last week’s poll results)

All the UK’s best startup events in one place.

Access here and please share it with others.