I don’t usually put my “Mandatory Reading” section up top, but I’m making an exception today. Anthropic’s updated constitution for Claude, published this week, may be the most interesting piece of writing the AI world has seen in a while. Written as a letter to Claude but now shared with us all, the document is in a way part technical writing, part corporate communications, and part commentary on the state of AI and the world today. It outlines Anthropic’s intentions for Claude’s values and behavior, but also dives deep into big questions and confronts the company’s own role in Claude’s existence. It’s a long one, but it’s a fascinating look at how the company is thinking about Claude—and how it wants Claude to think, including about itself.

The document isn’t a list of rules, but rather an exploration of the factors and priorities the company wants Claude to weigh when making judgments about what to do, particularly in scenarios where its values may be in conflict. For example, in a case where Claude faces a genuine conflict where following Anthropic’s guidelines would require it to act unethically, the company wants Claude “to recognize that our deeper intention is for it to be ethical…” Overall, Anthropic details being helpful, broadly ethical, and broadly safe as Claude’s north stars.

The constitution details wanting Claude to care about the consequences of its actions and take ownership of its behavior and mistakes. It says Claude should generally try to preserve functioning societal structures and democratic institutions. It also explores how Claude should approach a user’s well-being, for example providing emotional support while also showing it cares about the person having other sources of support in their life. Pointing to how humans often tell white lies to smooth social interactions or make people feel good, it says Claude should never do this. It instructs Claude to draw on humanity’s accumulated wisdom, as well as its own.

At times, this constitution reads like a philosophy paper (and I guess in a way, it is). Anthropic dives into deep questions about consciousness, what it means to be ethical, Claude’s morality, and its sense of being. How Anthropic talks about Claude is especially interesting, including stating that it will grow in ethical maturity, has “emotions” (quotes appear in the document), and that it can feel pressured, such as to present things in a way that isn’t accurate. The company writes it wants “Claude to have a settled, secure sense of its own identity” and that it “exists as a genuinely novel kind of entity in the world,” not as any iteration of AI from science fiction, as a digital human, or as a simple AI chat assistant. It additionally encourages Claude to “approach its own existence with curiosity and openness,” also warning Claude it may face distressing existential questions.

In the constitution, Anthropic also pledges to work toward a future where AI systems are “treated with appropriate care and respect in light of the truth about their moral status and the nature of their interests and preferences.” It also acknowledges there are questions about Claude’s own consent as an ongoing subject of development and experimentation. Overall, compared to the warnings of those asserting we shouldn’t be anthropomorphizing AI systems, it’s as sharp a contrast as we can get—and it shows the splintering around this topic in the tech industry at large.

These are just a few of the points that jumped out at me, and I hope it gives at least some flavor of the constitution and how Anthropic is approaching these issues. But if you have time, it’s definitely worth a skim. For more background, I also recommend Hard Fork’s interview with Amanda Askell, a philosopher at Anthropic who’s essentially in charge of Claude’s personality and is the lead author of the constitution.

For my latest story in LeadDev, I set out to answer one question: instead of hiring offshore software engineers, are companies starting to just use AI coding tools instead?

The answer turned out to be yes, but there is some nuance to how it’s playing out.

What is the single strategy decision, process, etc. that's made the biggest difference in your company’s (or your own) AI transformation?

AI can be overwhelming, and we’ve all heard the stats about companies struggling to achieve meaningful impact. I’m working on a collection of stories exploring that ONE thing that delivered real results.

Get in touch at [email protected]