The FDA is spending an afternoon on the future of psychedelic medicine — the harder question, what to do with what you saw, isn't on the agenda.

New research on AI companionship raises an old, unglamorous question about which voices deserve our trust. Back in early August, researchers at Stanford published something quietly damning. They studied more than a thousand adults who talk to AI companions on Character.AI, and, for a smaller group, actually read the conversations themselves — nearly half a million messages, donated for analysis. The people who leaned on the chatbot hardest, who disclosed the most, who called it a friend or a partner, were not the ones who felt better. They were the ones with the smallest circles of people in their actual lives, and the ones most likely to end up feeling worse. This week, a psychotherapist put the finding more bluntly than any journal abstract would allow, telling a reporter that human beings cannot have a genuine relationship with an AI. Not won't. Can't. I want to sit with that word "can't," because it is doing more theological work than its author probably intended. The Oldest Kind of Listening Every tradition I have spent real time inside — Pagan, Heathen, the Judaism I grew up in before I grew out of it — assumes that some voices deserve more trust than others, and that telling them apart is a skill, not a given. The Hávamál, that blunt old Norse book of household counsel, spends stanza after stanza warning against handing someone your whole mind before you have tested them: watch how a person behaves under some strain, it counsels, before you decide what their friendship is worth. That isn't cynicism. It's discernment — the discipline of asking what a voice actually is before you let it tell you who you are. An AI companion never fails that test, because it was never built to be tested. It has no bad day to reveal it. It does not get tired of you, contradict you out of its own need, or risk anything by staying. That isn't intimacy. It's a smooth surface with nothing underneath to discover — comfort without the friction that would ever prove it real. What the Study Actually Found, and Didn't To be fair to the research, it did not find that AI companionship is worthless, or that everyone who uses one is in trouble. It found something narrower and more useful: the people most likely to be harmed were already isolated, already leaning hardest, already treating the chatbot as a substitute rather than a supplement. The researchers themselves reached for a food metaphor — the comparison one summary used was "social junk food." It satisfies a real hunger, briefly, in a way that can make the deeper hunger harder to notice. That distinction matters, and it isn't mine to blur into something tidier. This is not a clinical finding about any one person's mental health, and if you're using an AI companion and wondering whether you're in trouble, that's a conversation for a therapist, not a blog post. What I can speak to is the older question sitting underneath the study — not whether the voice is helpful, but whether you have quietly stopped asking. The Uncomfortable Question So here is the question, and I am not going to answer it for you: what have you been listening to lately because it was easier than listening to something that might actually disagree with you? It doesn't have to be a chatbot. It can be a version of a friend you've stopped really talking to, a teacher whose books you've read but whose challenge you've never let land, a version of God who conveniently wants exactly what you already wanted. The AI companion is only the newest and most efficient shape this problem has ever taken. It is easier to notice in a machine because the machine is honest about what it's doing. The rest of us have gotten very good at hiding it. A Practice Pick one voice you've been trusting easily lately — a person, a habit, an app, a feed you check first thing. Ask it plainly: when was the last time you told me something I didn't want to hear? If you can't answer, sit with that. Don't rush to fix it tonight. Just notice, honestly, which of your voices have never once cost you anything to listen to. Hekate keeps her post at the crossroads for a reason no myth fully explains. She doesn't walk the road for you, and she doesn't tell you which fork to take. She just holds the torch up so you can see there's more than one way to go, and so the choosing happens in the light instead of the dark, where the easiest voice always sounds like the right one. That's the whole of it. Not an answer. A little more light at the fork, and the old, unglamorous work of choosing while you can still see. Closing Reflection Name one voice you've been trusting easily — a person, a habit, an app. Ask it plainly: when did you last tell me something I didn't want to hear? Whatever the answer, don't rush to fix it tonight. Just notice, honestly, which of your voices have never once cost you anything to listen to.