Picture the world's most exclusive nightclub. Velvet rope, intimidating bouncer, a line that wraps around the block. Then one day the bouncer walks down the queue, taps exactly 100-plus people on the shoulder, and says "you, you, and you — come on in, but keep it quiet." Everyone else? Still standing in the cold, wondering what they did wrong.

That's roughly what just happened with Mythos 5, Anthropic's most capable AI system. On 26 June 2026, the US government partially reversed restrictions it had placed on the model over security concerns, clearing more than 100 vetted American organisations to access it again. Not the general public. Not every company that asked nicely. A curated guest list.

Here's the actual substance, stripped of nightclub analogies: Mythos 5 had been restricted from broader release amid concerns about its capabilities landing in the wrong hands (the kind of concerns that sound dramatic until you remember these models can genuinely do things that matter for national security, biosecurity, and cybersecurity). That restriction is now being loosened — but only for a vetted shortlist of American organisations the government has decided it can trust with the keys.

It's not a full reopening. It's a side door, opened selectively, for people who've already passed some kind of background check.

And that's the part worth sitting with: this isn't really a story about AI capability anymore. It's a story about governance. Who gets to decide which companies are trustworthy enough to use the most powerful tools on the planet? Right now, the answer is: a government agency, working off criteria most of us will never see, making calls that could meaningfully shape who wins and loses in entire industries.

If you're running a startup, a scale-up, or even a established SME, this matters more than it sounds like at first glance. We're watching the early formation of a new kind of competitive advantage — not capital, not talent, but clearance. Being on the approved list for frontier AI access could become as valuable as a patent or a key hire.

That creates an obvious problem: if access is selective and security-driven rather than market-driven, the companies that get in first aren't necessarily the best-positioned to use the technology — they're the ones who best satisfied a vetting process. (Translation: being good at your business and being good at passing a government security review are two very different skills, and right now only one of them gets you through the door.)

For founders and executives outside that list of 100-plus, the lesson isn't to panic — it's to pay attention. Government-gated access to the most powerful AI systems is a new variable in the competitive landscape, and it's not going away. If anything, expect more of these tiered, selectively-reversed restrictions as governments figure out, in real time, how nervous they should be about who's holding the most capable models.

There's also a bigger question lurking underneath, one that goes well beyond this single approval: should any single government get to play gatekeeper for technology this consequential? Or does that kind of concentrated decision-making power deserve its own scrutiny?

So here we are: the bouncer reopened the rope for a select few, the line outside got a little shorter but didn't disappear, and everyone left standing in the cold is now asking a much more interesting question than "can I get in?" They're asking "who decided you couldn't?" That's the debate that's actually just getting started.

— The Business Index Team