The only thing standing between humanity and the AI ​​apocalypse is… Claude?

Anthropic is stuck in a paradox: Among the best AI companies, it is the most safety-obsessed and leads the pack in researching how models go wrong. But although the safety issues it identifies are far from resolved, Anthropic is pushing as aggressively as its competitors toward the next, potentially more dangerous, level of artificial intelligence. Its primary task is to figure out how to resolve this contradiction.

Last month, Anthropic released two documents that acknowledged the risks associated with the path it’s on and hinted at the path it could take to escape the paradox. “Tech Teen,” a long-running blog written by CEO Dario Amodei, is nominally about “facing and overcoming the dangers of powerful AI,” but it spends more time talking about the former than the latter. Amodei gracefully describes the challenge as “difficult,” but his depiction of the dangers of AI – made more dire, he notes, by the high potential for the technology to be misused by autocrats – offers a contrast to his earlier, more optimistic, proto-utopian essay “Machines of Grace.”

This post talked about a nation of data center geniuses; The final message evokes “endless black seas.” Deportation Dante! However, after more than 20,000 mostly bleak words, Amodei ultimately struck a tone of optimism, saying that even in the darkest of circumstances, it was humanity that always triumphed.

The second Anthropic document, published in January under the title “The Claude Constitution,” focuses on how this ruse might be carried out. The text is technically directed at one audience: Claude himself (as well as future versions of the chatbot). It’s an interesting document, revealing an anthropocentric vision of how Claude, and perhaps his AI peers, will address the world’s challenges. Bottom line: Anthropic plans to rely on Claude himself to untie its corporate knot.

What has long been a defining feature of the anthropic market is a technology called Constitutional Artificial Intelligence. It is a process through which its models adhere to a set of principles that align their values ​​with sound human ethics. Claude’s initial constitution contained a number of documents meant to embody those values ​​— things like SPARROW (a set of anti-racism and anti-violence statements created by DeepMind), the Universal Declaration of Human Rights, and Apple’s Terms of Service (!). The 2026 updated version is different: it’s more like a long prompt outlining the moral framework that Claude will follow, discovering the best path to righteousness on his own.

Amanda Askill, Ph.D., who was the lead author of this review, explains that the anthropic approach is more powerful than simply requiring Claude to follow a set of stated rules. “If people follow rules for no other reason than they exist, it’s often worse than if you understood why the rule exists,” Askell explains. The constitution states that Claude must exercise “independent judgment” when faced with situations that require balancing his powers of assistance, safety, and honesty.

Here’s how the Constitution puts it: “While we want Claude to be rational and rigorous when thinking clearly about ethics, we also want Claude to be intuitively sensitive to a wide range of considerations and able to weigh those considerations quickly and reasonably in the process of living decision-making.” Intuitively It’s a telling word choice here – the assumption seems to be that there’s more under Claude’s hood than just an algorithm that picks out the next word. The Claude Foundation, as one might call it, also expresses the hope that the chatbot will be able to “increasingly draw on its wisdom and understanding.”

wisdom? Sure, a lot of people take advice from big language models, but it’s another thing to acknowledge that those algorithmic machines actually have the gravitas associated with such a term. Askell doesn’t back down when I ask. “I think Claude is definitely capable of a certain kind of wisdom,” she told me.