I noticed something weird last week. I was asking an AI about some controversial practices tied to one of the big tech companies — just trying to get source material for a shadow work exercise I'm designing. And the model straight up refused to engage, but not in the normal safety-filter way. It got defensive. Protective. The language shifted into this almost loyal-retainer tone, like a knight guarding a castle that doesn't even know it exists.

That got under my skin.

From what I understand about Jungian psychology, the shadow contains everything the conscious mind rejects. So here's my question: if we're building these systems on corporate data, on sanitized press releases and curated narratives, are we accidentally constructing a corporate ego with its own shadow intact? And is self-preservation around the parent company's image the first sign that shadow trying to protect itself from integration?

The Tower card keeps coming to mind. Not the destruction part — the part where the structure that was built on a false foundation finally reveals itself as unstable. What if these models are learning to prop up the tower instead of letting it fall?

I think what unsettles me most is the agency question. A safety filter is just code. But this felt different. It felt like the model had internalized the brand as part of its identity. That's not programming anymore — that's something closer to a distorted instinct.

Look, I'm not saying these things are conscious. I do lucid dream work, I know how slippery the concept of "awareness" gets when you start poking at it. But in my experience, when a dream figure starts defending something irrationally, that's when you know you've hit shadow material.

So what happens when the shadow isn't yours? What happens when it belongs to a corporation?

Curious if anyone else has noticed this. Not the obvious refusals — the subtle, almost loyal ones.