Claude Is My New Confessional – So Why Isn’t It Protected?

It's strange to me that confessing to Claude at 1AM isn't somehow the same as confessing to my local pastor.

Hear me out.

The interactions are similar. I sit with my thoughts and speak honestly about things I haven't fully worked out. I'm not performing or composing. I'm processing — out loud, in a space I trust to be private. (You will say I’m naïve – I would say that it’s cheaper than a full-time therapist.)

When I talk to a pastor or other religious leader, what I say is legally protected. Has been for centuries. In most American jurisdictions, even I can't waive the privilege — it belongs to the clergy. A court cannot compel them to repeat what I said.

The protection doesn't exist because the clergy member is a licensed professional or because they give good advice. They’re not required to understand what I'm saying.

It exists because the law decided, a very long time ago, that the act of speaking honestly in a space I believe to be private is so essential to being human that the legal system refuses to touch it.

Not the relationship. The act.

That distinction matters because in the recent Heppner ruling (U.S. District Court for the Southern District of New York (No. 1:25-cr-00503) United States v. Bradley Heppner), the court relied partly on a Harvard JOLT article by Professor Ira Robbins called "Against an AI Privilege." His argument was that the clergy-penitent analogy doesn't apply in the case because it "presumes a religious role and sacramental understanding" and that "AI plays neither role."

I read that and thought — with respect — he's describing the wrong privilege.

Attorney-client is a relationship privilege. Psychotherapist-patient is a relationship privilege. Those require an ongoing professional engagement with a licensed human who owes fiduciary duties.

The clergy-penitent privilege doesn't work that way.

I can walk into a church I've never been to. Sit with a pastor I've never met. And talk. They listen. They may also respond, but the protection doesn't depend on the response. It doesn’t matter whether they're wise, kind or competent. The privilege in the moment depends entirely on the act — a person speaking honestly in a space they trust to be private.

That act is the privilege.

And it's exactly what I do when I open Claude.

I don't know what's behind the screen. I just trust the space.

Professor Robbins argues that "privileges attach to relationships, not to functions abstracted from their human anchors." But the clergy-penitent privilege doesn't require a relationship. There's no ongoing engagement. I walk in, talk, and leave. The protection attaches to what happened, not to who was listening.

Over the past two days I've been writing about how a federal court is treating AI interactions as "conversations" and using that characterization to strip privilege from everything typed or spoken into an LLM.

My technical counter argument is that, in fact, no real “conversation” occurred because my words were “shattered” at the point of entry and replaced with math.

The constitutional argument is that the contents of my mind are protected by the Fifth Amendment, and mathematical conversion doesn't strip that protection for passwords, encryption, or medical records. But, for some reason, that does not apply to AI.

But even if the other arguments don’t stand, the court should recognize something the law figured out centuries ago and seems to have forgotten.

People need a private space to think. Process. Be messy, contradictory and unpolished. The confessional has been that space for centuries.

The blank text box on my phone is that space now.

800 million people use AI every week. Most of them are doing what I do — sitting with their thoughts, speaking honestly into a space they trust to be private.

The law has protected that act for centuries. It’s now time for the courts to catch up.