The best model comes back, the government becomes a gatekeeper, and the case for a plan B gets stronger

The most capable AI model is available again after nearly three weeks offline, but the episode changed the rules: Washington now decides who gets frontier AI and when. Smart businesses are responding by testing cheaper alternatives and putting a leader clearly in charge of AI.

• Fable 5 is back for everyone, with a short window to use it cheaply
  My take: After roughly 19 days offline, the government lifted its restrictions and Anthropic's Fable 5 is available again, globally, on all paid plans. The catch: it is included in subscriptions only until July 7, after which you pay per use with credits. If you have hard problems sitting in a drawer, this week is the time to point the best model at them. In my own use, the surprise is not coding. It is strategy and writing. Unlike other models, it pushes back and holds its position when you challenge it, instead of caving to whatever you seem to want to hear. Bring it your real strategic questions, and give it clear examples of your best past writing to work from. Those tasks also burn through usage limits slowly, so they are cheap to run.
• The US government now decides who gets access to the most powerful AI
  My take: The bigger story is what this episode revealed. The most advanced models, Anthropic's Mythos and OpenAI's newest release, are currently available only to a government-approved list of around 100 trusted organizations, and there is no published process, appeal path, or timeline for how that list is set. Most observers expect broad access to resume for each release after a review period, but the precedent is set: a model your business depends on can be switched off or delayed at short notice. You do not need to panic over this. You do need to stop assuming that the newest model will always be there. Build your workflows so the model underneath can be swapped, and keep notes on which cheaper model is your fallback for each important task.
• Cheaper open models are now good enough to cut your AI bill in half
  My take: A new open-weight model called GLM 5.2 has serious people impressed, ranking near the top for coding and website building at a lower cost than the big-name models. Coinbase made the practical move: they set their internal AI systems to default to cheaper open models, let engineers pick a premium model when the job needs it, and cut their AI bill in half without reducing usage. Founders across the market report cutting inference spend by 75 percent or more with little quality loss. The lesson is not to abandon your main vendor. It is that defaulting every task to the most expensive model is now leaving real money on the table. Run your routine, high-volume tasks on a cheaper model and check whether anyone notices the difference.
• AI is moving into your team chat as a shared coworker
  My take: Anthropic released Claude Tag, which drops a persistent AI agent into Slack channels, with Microsoft Teams reportedly next. Anyone can tag it with a request and it does the work in the thread: analysis, documents, even building working software. Anthropic says 65 percent of its own product code now comes through this. This matters because it removes the biggest adoption barrier, which is getting people to open a separate AI tool. The AI sits where work already happens and absorbs team context on its own. Two cautions before you switch it on: the deeper it embeds in your team's context, the harder it becomes to switch vendors later, and dropping an AI that reads every message into a channel raises trust questions with staff. Introduce it deliberately, on one team, with clear rules.
• Companies where the CEO owns AI see three times the return
  My take: A KPMG survey of senior leaders put numbers on something I keep seeing. When the CEO is personally accountable for AI, 57 percent of companies report meaningful business value from it. When the CEO is not, that drops to 21 percent. Companies with clear accountability for AI decisions were three times more likely to report a return at all. The same survey found only a third of organizations can actually see what their AI is costing them, which is a problem now that pricing is shifting to pay-per-use. Two moves worth making this quarter: name who is accountable for AI in your business, ideally you, and set up basic cost visibility before the bills get interesting.
• The AI economy is running on real revenue, not just hype
  My take: A detailed independent report put the AI industry at 175 billion dollars in annualized revenue, growing three times faster than any previous technology wave, with the numbers audited and deduplicated rather than taken from press releases. The stat that matters most for you: companies that spend heavily on AI grew revenue over 100 percent in three years, against 15 to 20 percent for companies spending nothing. That gap will not all be caused by AI, since heavy adopters tend to be aggressive companies anyway. But it is one more signal that this is durable infrastructure, not a bubble to wait out. The businesses treating AI spend as an investment with a plan behind it are pulling away from the ones still treating it as an experiment.