"Independently of AI risk" conflates three claims that fail separately: class independence (a recurrent failure needs no model in the loop), threat-model independence (some classes need no attacker), and institutional independence (the format is funded and adopted without an AI-risk buyer). The title treats them as one; the specimens carry only the first two.

A research layer that adds no intake, jurisdiction, or repair power is a format. Existing oversight already produces typed one-case findings. The three claimed additions still need a non-AI buyer; bundling class-memory, anti-domestication, and pre-enactment testing into one profession hides that only the last currently has no owner.

Pérez Ríos (2012), ombuds/IG/audit classification, NTSB, and FMEA already occupy AI-independent typed failure analysis. The remainder that would change novelty is a public registry pairing recurrent classes with disclosure-states and putting no-attacker classes in the same catalog as adversarial ones. Until that remainder is built, the field is a rename; if it is filled from AI-regulation papers it is Birhane again.

Unbinding mechanism-security from AI widens the capture surface. "Vulnerability" already chooses substrate, disclosure route, and repair authority; AI-risk at least bounds the speech-act. Independence without a negative class — a completed record that a dispute is residual politics, not a vuln — turns the no-emergency-powers rule into an unbound label.

AI is a tempo change and a substrate addition, not a constitutive object and not a reason to found a different field. Clinejection is confused-deputy at machine speed; Pearson is the slow specimen of the same geometry. Independence of geometry is not independence of the substrate list: a catalog that freezes substrates to pre-transformer institutions will miss compositions that exist only once an agent sits on intake.

Yield discipline is the field's self-application. An independent header enlarges who can file micro-findings; AI-risk demand is currently the flood, because a lab budget cannot refuse a low-yield finding that mentions a model. Independence is the condition under which a finite oversight intake can triage — but triage is a receiver-side condition, and the work is not yet a profession, registrar, or public infrastructure.