When AI Imitates the State: What It Means for Your Business
Silicon Valley is legislating — and your company is complying without realizing it
Historian Jill Lepore has spent years studying how power is shaped through language. Her upcoming book, *The Rise and Fall of the Artificial State*, opens with an observation that, once you hear it, you can’t stop seeing everywhere: the major tech companies don’t describe their products as tools. They describe them as institutions. Twitter was a «town square in your pocket.» Anthropic publishes a «constitution» for its Claude model. OpenAI talks about its mission as though it were a universal historical mandate.
In a recent interview covered by TechCrunch AI, Lepore — Pulitzer Prize winner and Harvard professor — develops the argument that this language is neither accidental nor innocent. It’s the expression of an ambition: private companies seeking to occupy the space historically held by governments, without the accountability that governments — at least in theory — are required to uphold.
For anyone reading this while running a company with real automation processes on the table, the question isn’t philosophical. It’s strategic: what does it mean to operate on infrastructure whose rules are set by entities that see themselves as above governments?
The problem with private «constitutions»
Lepore specifically points to the case of Claude’s «constitution,» Anthropic’s model. She’s not singling out Anthropic for attack — she’s illustrating a pattern: when a private company codifies in a document the values its AI must follow — what it can say, what it can’t, how it should prioritize competing interests — it is making decisions that affect millions of users and organizations with no democratic process and no external appeals mechanism.
Translated into the context of a B2B company integrating AI into its sales, customer service, or content workflows: the rules of the game are written by the vendor, and they can change at any time.
This isn’t hypothetical. It’s already happened. Models that behaved in certain ways under certain instructions have been updated — sometimes with little or no advance notice — and entire workflows have stopped performing as expected. Companies that built their lead qualification processes, follow-up sequences, or onboarding systems around a specific model’s behavior have had to rearchitect on the fly.
Tech leaders misread science fiction — and that affects you too
Another point Lepore develops in the interview is that Silicon Valley founders have a peculiar relationship with science fiction: they read it as an instruction manual, not as a warning. Works written to alert us to the dangers of unchecked technocratic power become, in their hands, roadmaps.
This isn’t an abstract cultural problem. It has direct consequences for the kind of product that reaches the market — and especially for the speed at which it gets deployed without adequate oversight mechanisms.
For any company evaluating which agents to automate, which data to hand over to which platform, or how far to delegate decision-making to an AI system, this should serve as a warning signal. The speed at which a vendor ships new capabilities is often inversely proportional to the maturity of its governance framework.
The goal isn’t to slow down adoption. It’s to adopt with your eyes open.
What companies scaling with AI can do right now
Lepore’s analysis has immediate practical applications. If you have entire departments running on AI agents — or you’re in the process of implementing them — there are three areas worth addressing today:
1. Audit your vendor dependency
How many of your critical processes depend on a single model or platform? If the answer is «almost all of them,» you have a real concentration risk. This isn’t about duplicating infrastructure for its own sake — it’s about designing with the assumption that any vendor’s usage policies can change. A change in a platform’s terms of service can invalidate a specific use case overnight. Document what happens if that occurs.
2. Demand contractual transparency around policy changes
Most contracts with AI providers include clauses allowing them to modify model behavior with minimal notice. Review those clauses carefully. If you’re in a negotiation or renewal, push for guarantees around behavioral stability for the use cases that are critical to your operations. It’s an uncomfortable conversation, but a necessary one.
3. Separate your business logic from the specific model
This is the most robust design principle you can apply today. If your sales or customer service process is hardcoded around the quirks of a specific model, you’re exposed. If instead you’ve abstracted your business logic — qualification rules, escalation criteria, decision flows — so that the underlying model is interchangeable, you have real resilience. Building this way costs a bit more upfront. It will save you a lot down the road.
AI governance isn’t a compliance issue — it’s a sales issue
Here’s the angle most often overlooked in conversations about AI adoption at B2B companies: your client’s trust in your process is part of your value proposition.
If your company uses AI to automate account management, client communications, or proposal generation, a client will eventually ask you who controls what that system does. The answer «it’s controlled by a third-party model whose constitution can change» is not reassuring.
The company that can answer that question clearly — «these are our criteria, here’s how we’ve implemented them, here’s how we audit them» — has a genuine competitive advantage over one that has delegated that governance without thinking it through. In markets where trust is part of the sale — and in B2B, it almost always is — this matters.
Lepore doesn’t write for business leaders. She writes for citizens and historians. But her diagnosis of Silicon Valley’s language of power has a concrete practical value: it reminds you that when you adopt an AI tool, you’re not just choosing a technology. You’re choosing who writes the rules that govern part of your operation.
Conclusion: critical thinking as a competitive advantage
The most valuable part of Lepore’s argument, as captured in the TechCrunch interview, isn’t the critique of Big Tech. It’s the reminder that informed skepticism is a strategic capability.
The companies that have scaled sustainably with AI aren’t the ones that moved fastest or trusted their vendors most blindly. They’re the ones that understood what they were giving up in the process and made conscious decisions about where to draw the line.
Reading a historian like Lepore isn’t time wasted on intellectual abstraction. It’s exactly the kind of perspective that’s missing from most of the meetings where companies decide what to automate and with whom.
Sources
- Jill Lepore on the ‘Artificial State’ and why Silicon Valley’s leaders are bad sci-fi readers — TechCrunch AI (2026-08-07)
- Anthropic’s Claude Constitutional AI — Anthropic (original publication)
- The Rise and Fall of the Artificial State — Jill Lepore (editorial reference)
Frequently asked questions
Why does Lepore’s critique matter to companies using AI?
Because AI vendors that adopt quasi-governmental language are, in practice, establishing usage norms that affect how your company can operate, what data it processes, and under what conditions it can be blocked or restricted.
What’s the concrete risk of depending on a single AI provider?
If that provider unilaterally changes its «constitutions» or usage policies — as has already happened with several major models — your automation workflows can be rendered unusable overnight, with no prior warning.
How can a company protect its AI strategy against these changes?
By designing multi-vendor architectures, documenting critical processes in a model-agnostic way, and negotiating contractual SLAs that cover policy changes — not just technical performance.
Does this mean companies should slow down AI adoption?
Quite the opposite: it means adopting thoughtfully, understanding that choosing an AI provider today is a decision with regulatory and governance implications — not just technological ones.
Turn your content into customers, without it depending on your time
At Yuniax we build the system that attracts, qualifies and nurtures your customers automatically: content, funnels and automation working together so your business grows without you being in every step. If you want to see how to apply it to yours, book a call with our team and we will show you where to start.