The biggest AI risk for your business in 2026 might not be falling behind on capability; it might be who controls your access to it.
Over the last fortnight we have seen something genuinely new. The US government has started deciding who can use the most powerful AI models, customer by customer.
First, the capability, because it deserves to be said. Cisco recently pointed Anthropic's most advanced model, Fable/Mythos, at 1.8 billion lines of code and reviewed it in eight weeks. Their estimate for the traditional way, humans plus existing tooling, was over eight years. Now followed by OpenAI's new model: GPT-5.6 is its strongest model yet across coding, biology and cybersecurity. The pace is extraordinary, and still accelerating.
But here's what changed. On 12 June, the US Commerce Department forced Anthropic to pull both Mythos and Fable for every customer, worldwide, overnight. Last week it partially reversed, but kept the keys. A short list of approved American companies now gets access, and the government decides who is on that list. Days later, OpenAI released GPT-5.6 to around 20 government-vetted partners.
I keep coming back to one implication.
If your business plans to build on the genuine frontier, your supply chain now runs through Washington, not a vendor contract you can renegotiate. A government approval queue, with discretion to slow a release, decide who qualifies, or switch a model off mid-contract. That is supply chain risk, and very few strategy decks have a line for it.
Does open source now deserve a serious seat at the table, and not as the budget option?
The leading open-weight models have closed most of the gap. Z.ai's GLM 5.2 beats GPT-5.5 on several long-horizon coding benchmarks at roughly a sixth of the cost, and sits within a point or two of the closed flagships on real agentic work. You can run the weights on your own infrastructure, keep your data in house, and nobody can revoke access by letter or surprise you with a usage bill. For a lot of genuine business work, "good enough and fully owned" quietly beats "frontier and gated".
But this is where I do not have a settled view, and I would value yours.
Four of the top five open models today come from Chinese labs. So choosing open source does not remove dependency, it swaps one kind for another. You step out of the Washington queue, but take on geopolitical exposure, licensing fine print, and full responsibility for safety and security.
So the question I am sitting with is this: does building on open source reduce your risk, or quietly increase it?
And underneath it, the question every leadership team should be asking. Is your AI advantage something you rent from a frontier lab, or something you own? Because the companies that get this right will not just be the ones with access to the best model. They will be the ones who understood, early, who controlled it. Where do you land?