I have a confession: I have a minor obsession with niche perfume. The kind with names like "Fat Electrician" and quirky fragrance notes. What fascinates me isn't just the weird smells, it's what perfume has to say about technology, selfhood, and what resists digital control.

At some point, assistance becomes agency. Maybe the question isn’t whether AI “thinks,” but what happens when it starts taking the next steps without us.

Working on device trust taught me how much implied belief is baked into the world. You can’t verify every layer; at some point you stop checking and start believing. AI is breaking the conditions that made that bargain possible.

Safety-critical domains treat deployment as part of the risk surface. Why does AI safety still feel so model-centric?