Who Gets the Best AI Now: Governments Start Choosing

The thread running through this fortnight is control. In Washington, the government has started deciding who may use the most capable AI models before they reach the open market, and it applied that power to both OpenAI and Anthropic within the same two weeks. The United Nations opened its first summit devoted entirely to governing the technology. Away from the policy noise, China is leading the part of AI most likely to reach your factory floor, while fresh evidence suggests the enterprise excitement of the past two years has not yet paid for itself. For a manufacturing leader, the useful question is where these shifts touch your access to tools and the return you get from them.

1. The UN Opens Its First Summit Built to Govern AI

The UN’s Global Dialogue on AI Governance opened in Geneva on 6 July, running to 7 July, as the first standing UN platform created solely to set rules for the technology. It follows the first report from a new UN scientific panel of 40 experts, which warned that science cannot yet guarantee advanced AI will avoid catastrophic harm, and that frontier development is concentrated in the US and China. Participants pointed to a widening gap between countries with strong AI infrastructure and those still short of basic connectivity.

What this means for you: Global rules for AI are being drafted now, and they will eventually shape what you can sell, where, and with what disclosures attached. The concentration of frontier development in two countries means most firms elsewhere will use tools built under someone else’s rules. India sits between these blocs, which affects both the models you can reach and the compliance you will face.

Action tip: Track AI regulation the way you track a tax or trade change, because it will carry that level of consequence. Give someone the job of watching the EU AI Act timeline and other emerging frameworks, and logging where your products or content would need disclosure or documentation. Building that record now costs far less than reconstructing it against a deadline.

2. OpenAI Ships New Models, but Washington Chooses Who Gets Them First

OpenAI released its GPT-5.6 family, named Sol, Terra and Luna, then limited early access to a small group of partners whose names it shared with the US government. The restriction followed a recent executive order asking frontier developers to submit powerful models for government review up to 30 days before release. OpenAI complied, but said publicly that government-gated access should not become the default because it keeps strong tools from the businesses and developers who need them.

What this means for you: A government checkpoint now sits between a frontier model’s launch and your ability to use it. Access to the most capable tools may arrive later for buyers outside a vetted circle, and later still for those outside the US. Pricing is tiering sharply as well, with the flagship model costing several times more per use than the lighter versions.

Action tip: Do not design a critical workflow around a single model you assume will be available on launch day. Build for the tier you can actually access, usually the mid or low-cost option, and confirm it clears your task at an acceptable cost before you commit. Keep a second provider qualified so a delayed or restricted release does not stall you.

3. Anthropic’s Most Powerful Models Were Pulled, Then Partly Restored

Anthropic released Claude Fable 5 and the more restricted Mythos 5 in early June, then had to disable both after the US issued an export-control directive suspending foreign access on national security grounds. The shutdown briefly cut off paying customers worldwide, not only foreign users. By early July access had been restored, with the government clearing the models for a set of vetted partners. Anthropic has set out its own account of the events in a public statement.

What this means for you: A tool your business depends on can be switched off by a government with little notice, and the reason may never be fully explained to you. Capability itself is now a trigger for restriction, so the most advanced model is also the one most exposed to sudden limits. This is a supply risk as much as a technology choice.

Action tip: Treat your AI provider the way you treat a critical single-source supplier. Keep your prompts, data and integration logic portable enough to move to another model within days. Ask vendors directly how they would handle a forced suspension, and weigh the answer into any contract for something business-critical.

4. China Is Winning the Robotic Hands That Will Reach Factories First

Chinese firms now lead the market for dexterous robotic hands, the multi-fingered grippers that let humanoid robots handle real objects. Companies such as LinkerBot ship more than 1,000 high-articulation hands a month, at prices near 900 dollars against 150,000 dollars or more for comparable Western units. Tactile sensor costs in China have fallen by more than 99% in a few years. Early factory trials of humanoid robots report 30% to 50% of human productivity on narrow, structured tasks.

What this means for you: The hardware that makes a robot useful on a production line is becoming cheap and available at volume, and it is being built mainly in China. For a manufacturer, physical automation is shifting from a distant prospect towards a cost decision within a few years. The current performance ceiling is real, so these robots suit repetitive, well-defined work rather than variable tasks.

Action tip: Identify the two or three most repetitive manual tasks on your line, the ones with stable inputs and clear steps. Price them against the falling cost of robotic manipulation on a three-year horizon rather than today’s figures. Following Chinese suppliers now gives you a cost benchmark before the technology reaches your competitors.

5. The AI Hangover Has Arrived, and the Numbers Explain Why

The industry mood has turned from excitement to a reckoning over results, a shift now widely called the AI hangover. An MIT study found that 95% of enterprise generative AI pilots delivered no measurable impact on profit and loss, with the failure tied to poor integration rather than weak models. The same study found that bought, workflow-specific tools succeeded far more often than systems companies built alone. One manufacturing executive told the researchers that little had actually changed in daily operations, whatever the online enthusiasm suggested.

What this means for you: The gap between AI’s promise and its measured return is now documented, which gives you cover to demand proof before you spend. Most failures came from tools bolted on without changing how the work actually runs. The projects that paid back were narrow, deeply embedded, and often bought from a specialist rather than built in-house.

Action tip: Pick one specific, repetitive process and set a single financial target for it, such as hours saved or cost removed, before you begin. Favour a focused external tool that fits your workflow over an ambitious internal build. A defined operating layer such as AIFM exists to hold this discipline, turning scattered pilots into a small number of measured, working systems.

6. AI Readiness Is Not a One-Time Checklist, a New UN Report Argues

A UNDP report published on 6 July, drawing on 26 country assessments, argues that AI readiness changes meaning once adoption is already underway. It finds that AI enters public and business systems through procurement, vendor platforms and infrastructure choices, before any formal strategy is in place. At that point readiness becomes a running diagnostic, used to spot where dependencies and risks are forming, rather than a box ticked before launch.

What this means for you: You are probably already adopting AI through the software you buy, not through a decision you consciously made. Each vendor that adds AI to a tool you use creates a dependency on their model, their pricing and their availability. Readiness is therefore about knowing where those dependencies sit, not about a single preparation exercise.

Action tip: List the everyday tools your business runs on and mark which have added AI features in the past year. For each, note what you would lose if that feature changed, rose in price or was withdrawn. This map shows your real exposure and tells you where to keep an alternative ready.

That’s All for This Fortnight

The pattern this fortnight is control meeting reality. In Washington, access to the strongest models has become something governments grant rather than markets sell. Elsewhere the effort is going into writing rules and gauging exposure to tools built by others. In practice the story is simpler, with China supplying the useful hardware while most enterprise pilots still fail to pay back. The firms that gain will be the ones that keep their options portable and spend only where the return is proven.

Until next time, The AI-First Mindset Team

About Us

At AI-First Mindset, we help leaders bridge the gap between knowing AI and using it. Our work includes hands-on workshops and operating frameworks designed for companies that want to embed AI into workflows without relying on large institutional vendors.

To explore how we can help, contact us at aifirstmindset.ai.