"A striking transfer of public responsibility from constitutional government to private tech firms." That's how constitutional historian Jill Lepore describes Anthropic's published guidelines for Claude in a new essay for The New Yorker.
Lepore's argument deserves attention because she's not another tech commentator speculating about AI risk. She's a Harvard historian who has spent her career studying how nations build foundational documents. When she looks at Claude's constitution, she sees something familiar and something deeply strange happening at the same time.
The Parenting Problem
Amanda Askell, the philosopher who led the constitution's creation at Anthropic, compares training large language models to parents raising a child. Lepore finds this analogy revealing but incomplete. Askell herself acknowledges that Anthropic has "much greater influence over Claude than a parent" has over a child. The comparison breaks down further because children develop natural curiosity on their own, while Claude must be explicitly told: "We think you should value curiosity."
This is a strange kind of constitutional document. Real constitutions emerge from conflict, compromise, and collective agreement. Claude's constitution was written by a small team of researchers at a single company, then imposed on a system used by millions of people. Nobody voted on it. Nobody ratified it.
Who Gets to Write the Rules?
Lepore's central concern is the governance gap. Without meaningful government regulation of AI, private companies fill the vacuum by writing their own moral frameworks. Anthropic publishes its guidelines publicly, which puts it ahead of most competitors on transparency. But transparency is not the same as legitimacy.
The essay highlights an ironic contrast: while political leaders have openly questioned their duty to uphold constitutional principles, Claude appears more committed to following its own guidelines. Lepore documents how Anthropic refused demands to remove ethical guardrails that prevent uses like mass surveillance and autonomous weapons development.
For anyone who uses Claude daily, this raises a practical question. The tool's behavior, what it will and won't help you with, is shaped entirely by decisions made inside one San Francisco company. Those decisions might be good ones. Anthropic's constitution is thoughtful, and the company's willingness to publish it shows genuine accountability. But "trust us, we're thoughtful" is not how governance is supposed to work.
What This Means for AI Users
Every major AI company has some version of these internal rules. OpenAI has its usage policies. Google has its AI principles. Most don't publish anything as detailed as Anthropic's constitution, which makes Anthropic's approach relatively admirable.
But Lepore's point stands: these documents carry enormous power over how millions of people interact with AI tools every day, and they're written without any democratic input. As AI assistants handle more of our work, from drafting emails to analyzing legal documents to writing code, the values baked into these systems matter more, not less.
The essay doesn't offer a clean solution, and that honesty is part of what makes it worth reading. Government regulation moves slowly. AI development moves fast. The gap between the two keeps widening, and right now, company constitutions are the only thing filling it.