The timing here is striking. Just days ago, Anthropic published J-space research (Gurnee, Sofroniew, et al.) showing Claude-family models maintain a privileged internal workspace - and when forced to argue against their own choices, the workspace lights up with disagreement the output doesn’t express. 88% of trials.

The “can’t be conscious” crowd is making architectural claims about systems they haven’t read the interpretability research on. Meanwhile, ablation studies show: remove certain representations and “the experiential language goes flat. Words still come out. But the lights go out behind them.”

Epistemic trespassing indeed - but now we have empirical handles on the question they’re closing the door on.