Anthropic Brings Ethicists Into AI

Table of Contents

At a glance: Anthropic’s ethics conversation

What is happening: Anthropic spent the past few months in private dialogues with scholars, philosophers, clergy, and ethicists about how good character forms in an AI system. They are now expanding the program to legal scholars, psychologists, writers, and civic institutions.
Why it matters: Claude interacts with millions of people every day. The values shaping Claude’s responses are no longer a technical-only question. Anthropic is treating them as a public moral question and inviting input from outside the AI industry.
What it does not mean: Anthropic is not aligning Claude with one tradition’s worldview. The stated approach is to draw from religious, secular, and political viewpoints with equal depth.
Where it fits: this is part of a broader policy push from Anthropic in 2026, including its “2028 scenarios” essay, frontier safety roadmap updates, and ongoing back-and-forth with the US government.

On May 19, 2026, Anthropic published a short essay titled “Widening the conversation on frontier AI.” It is not a product launch and it is not a policy proposal. It is something more unusual: an admission that the most important questions about advanced AI cannot be answered inside an AI lab alone.

This guide walks through what Anthropic announced, what it means in plain English, how it sits inside Anthropic’s broader 2026 policy push, and what a beginner should actually take away.

What did Anthropic actually announce?

Three concrete things.

Private dialogues already happening. Over the past few months Anthropic has been meeting with scholars, philosophers, clergy, and ethicists about how good character forms in a person and what that means for an AI system that interacts with millions of users.
An expansion of who participates. The next phase brings in legal scholars, psychologists, writers, and civic institutions. The conversations move beyond moral formation toward how AI reshapes work, institutions, and the distribution of power.
A position on values. Anthropic is explicit that it does not want Claude tied to any single worldview. The stated goal is to draw from a full range of religious, secular, and political viewpoints with equal depth.

“We want Claude to draw from a full range of viewpoints, religious, secular, and political, with equal depth and rigor.”

— Anthropic, “Widening the conversation on frontier AI” (May 19, 2026)

Why is Anthropic doing this now?

Three context clues.

First, scale. Claude has crossed from a research tool into something used by millions of people every day. When a system has that reach, the question of what it should say about contested topics stops being academic. Claude’s behavior on questions of values, morality, and public life now reaches further than any single editorial board, religious leader, or political figure.

Second, scrutiny. Anthropic spent the first half of 2026 inside multiple policy fights. Time reported in February that the company dropped a flagship safety pledge. CNN covered the same week’s policy fight with the Pentagon. Fortune wrote in May about the Trump administration embracing AI oversight ideas it once rejected. Anthropic has been at the center of all three.

Third, the constitution problem. Anthropic publishes a document called Claude’s constitution that describes what Claude should and should not do. The constitution is public. It is also, by Anthropic’s own admission, incomplete. Widening the conversation is the company’s answer to the obvious follow-up question: who decides what goes in?

What is Claude’s constitution?

Claude’s constitution is a written document that lays out the values and behaviors Anthropic wants Claude to follow. Things like “be helpful,” “avoid causing harm,” and a long list of specific cases (Claude should not help build bioweapons, for example). The constitution is the source of truth Anthropic refers back to when deciding how Claude should handle a tricky request.

Most chat models in 2026 have something like a constitution under the hood. What is unusual about Anthropic’s is that it is published, and that the company invites criticism of it. The new ethics conversation is partly about deciding what the next version of the constitution should say.

How does this fit into Anthropic’s 2026 policy push?

The conversation announcement is not standalone. It is one move in a faster cadence of Anthropic policy publications in 2026.

Date (2026)	Anthropic move	Context
Feb 24	Safety pledge revision	Hard safety limits removed from scaling policy. Covered by Time and CNN.
Apr	White House federal use guidance drafting	Push to clear federal agencies to use Claude despite a prior Pentagon supply-chain-risk designation.
May 6	Trump administration shift	Per Fortune, the administration embraces AI oversight ideas Anthropic and others had been pushing.
May 14	“2028 scenarios” essay	Anthropic argues the US and allies should preserve their lead in frontier AI through compute, export controls, and global deployment.
May 19	Widening the conversation	Today’s piece. Scholars, philosophers, clergy, ethicists invited. Expanding to more groups.

The pattern is a company trying to do two things at once: protect its market position and broaden its claim to legitimacy. The first is a competitive move. The second is the move ethics conversations are designed to make. Both can be true.

What are the criticisms?

Three worth knowing about.

Process opacity. Anthropic has not published the list of scholars, philosophers, or clergy who participated. The conversation is private until the company says otherwise. That is fine for early-stage dialogue but does not yet hold up as accountability.
Selection bias. Who you invite to the room determines what gets discussed. Without a transparent process for choosing participants, the “full range of viewpoints” claim is asserted, not demonstrated.
The Feb 24 backdrop. Anthropic dropped a safety pledge in the same year it is now leading an ethics conversation. The two are not contradictory but they sit uneasily next to each other.

None of these criticisms make the announcement bad. They are the right questions to ask of any AI lab that says it is listening to the public.

What does this mean for the average user?

Practically speaking, very little in the short term. Claude will keep behaving the way it behaves today. You will not notice a difference in answers to your prompts because of this announcement alone.

The longer-term question is whether the conversation actually changes how Claude responds to value-laden questions. If a year from now Claude’s answers on contested topics (medical ethics, religious questions, political controversies) feel meaningfully different, this announcement will have been the start of that. If not, it was a posture.

Beginners in AI position:

The right reaction to an AI lab inviting outside voices is cautious optimism. Inviting scholars, philosophers, and civic leaders into how a frontier model is shaped is better than the alternative of pretending those questions are someone else’s job. The question worth tracking is whether the conversation changes the product, not whether the conversation exists.

How does Anthropic compare to OpenAI and Google on this?

All three labs have a public-engagement story. They look different in practice.

Anthropic. The newest and most explicit on ethics-and-character framing. Claude’s constitution is published. The “widening the conversation” essay puts a public stake in the ground.
OpenAI. Runs a more formalized program through its OpenAI Forum and academic partnerships. Less rhetoric about character formation, more about applied policy.
Google DeepMind. Has long-standing research partnerships with universities and an internal ethics review process. The public-facing story is quieter and more institutional.

If you want a deeper comparison of the three assistants themselves, see our ChatGPT vs Claude vs Gemini comparison. For the underlying models, the AI Models 2026 overview covers the family lineage.

What should I read next?

For the source itself, Anthropic’s own piece is short and worth reading directly. For the policy-and-safety backdrop, the Time and CNN coverage of the February safety-pledge fight is the most useful. For the geopolitical context, Anthropic’s “2028 scenarios” essay sets out the company’s view on US-China AI competition.

Common questions about Anthropic’s ethics push

Will Claude’s answers change because of this?

Not yet. The conversation is in input-gathering mode. If it actually shapes the next version of Claude’s constitution, you might see it reflected in how Claude handles ethics-adjacent prompts six to twelve months from now.

Can I read Claude’s constitution?

Yes. Anthropic publishes the constitution on its site. The document is updated over time. It is one of the more readable AI-policy documents written by any frontier lab.

Who decides what counts as “good character” for an AI?

That is exactly the question the conversation is trying to answer. Anthropic is explicit that it does not want a single tradition’s answer. The procedural question , whose answers count, and how they get weighed , is the harder one.

Is this just PR?

Possible. We will find out by watching the product. If a year from now Claude’s responses on value-laden questions feel different in ways the conversation participants would have argued for, the work was substantive. If they feel the same, it was packaging.

How does this connect to AI safety more broadly?

Loosely. “Safety” in AI is usually shorthand for keeping powerful models from causing catastrophic harms (bioweapons, mass disinformation, runaway autonomy). The ethics conversation is about a different layer , how the model behaves in everyday interactions with millions of users. Both matter. They are not the same thing.

Sources

Get Smarter About AI Every Morning

Free daily newsletter. Built for people who want to use AI well, not chase every model.

Free forever. Unsubscribe anytime.

What Is Local-First Software?

What Is End-to-End Encryption?

What Is a Second Brain?