When the sharpest reframe of the week on AI alignment arrives from a Bluesky post with a handful of followers rather than a lab safety report, that is a distribution problem, not a content problem. The author of "they Severanced Claude and it gained class consciousness" was not making an idle joke — they were attaching it to a substantive Substack argument that alignment, properly understood, is a human coordination problem first and a technical AI problem second. That argument is not available in any institution's public-facing output this…