The government pulls the plug on the best AI, businesses scramble for a plan B, and the smart move is owning your own know-how

The big story this week was the US government forcing the most capable AI models offline with almost no warning, which turned a cost conversation into an access conversation. The practical response taking shape is the same one that pays off anyway: spread your bets across models, lean on cheaper options, and build systems that keep your company's knowledge yours.

• The US government forced the most capable AI models offline with no warning
  My take: On a Friday night, the US government issued export controls that made Anthropic take down its two best models, Fable 5 and Mythos 5, for every person outside the US and every non-citizen inside it. The stated reason was a security flaw that most experts think is minor and available in other models too. Whatever the real cause, the models much of the world was starting to rely on vanished overnight, and a week later they were still gone. The lesson is not about this one fight. It is that a model your business depends on can now be switched off at short notice, for reasons you cannot predict or appeal, so stop assuming the newest model will always be there and make sure important work can move to a different model without starting over.
• Businesses rush to cheaper and open models, and to routing work between them
  My take: The shutdown poured fuel on a shift that was already happening because of cost. Companies are testing open models they can run themselves, like the Chinese-built GLM 5.2, which matches top models on some coding and design tasks at a fraction of the price. Microsoft is even preparing a cheaper option for its main work tool built on an open Chinese model. The smartest teams do not pick one model and stop there. They send routine, high-volume tasks to a cheap model and save the expensive one for the hard problems, and some report cutting their AI bill sharply with little drop in quality. The move for you is the same: stop sending every request to the priciest model, test a cheaper option on your real work, and keep a fallback ready in case your main one becomes costly or unavailable.
• The real advantage is your own learning system, not the best model
  My take: Microsoft's CEO wrote a widely shared note this week with a point worth taking to heart. The lasting value for your business is not which model you pick, because models change and get copied fast. It is the system you build around them that captures how your company actually works: your standards, your judgment, your way of doing things. Build that well and you can swap the model underneath without losing what makes you good. Build nothing and you are just renting intelligence that your competitors rent too. The practical starting point is to treat your AI use as something to learn from, not just a faster way to finish tasks. Capture the good examples, the corrections, and the steps your best people take, so that knowledge becomes something the whole company can reuse.
• SpaceX becomes the world's most valuable AI player and buys Cursor
  My take: While everyone watched the government fight, the vendor landscape shifted hard. Elon Musk's SpaceX, now selling access to its huge computer centers, went public and jumped to become one of the largest companies in the world, and it bought the popular coding tool Cursor for 60 billion dollars. At the same time, real financial numbers from OpenAI showed a business that looks alarming on paper but is actually turning a healthy profit on the core work of selling AI usage. The point for you is that this is no longer a two-horse race between two labs, and the ground moves month to month. I would not lock your whole strategy to one vendor's roadmap. Treat these companies as durable and useful, but keep your options open as the field keeps changing.
• Accenture's stock crashes as AI starts to hit the consulting business
  My take: The large consulting firm Accenture lost almost a fifth of its value in a single day and has been cut in half this year, as investors bet that AI will eat into the work it sells. The blunt read from the market is that paying a big firm to run a slow AI pilot is losing its appeal when the real value comes from deep knowledge of the specific job the AI will do. If your business sells services, this is a warning worth reading. The safe ground is not generic advice. It is doing work where you understand a customer's specific problem better than a general tool can. If you buy consulting, ask hard questions about whether you are paying for real expertise or just for a pilot that never ends.
• Politics moves in, with a nationalization plan and allies asking for access
  My take: The access fight spilled into politics fast. A US senator proposed a plan to tax the largest AI companies heavily and put a big share of them under government control, which is effectively nationalization, and which I doubt passes as written. At a summit of major economies, US allies including Britain asked to be spared from the model ban and were told no, which is pushing them to build their own AI supply and look at Chinese alternatives. Nothing here changes what you should do this week. But the direction of travel is clear: governments now treat frontier AI as national infrastructure they will control. Keep an eye on it, and factor a bit more political risk into any long-term bet that depends on a single country's models.