For a week this brief tracked agent identity as a product category, then watched the card networks co-sign authorization as the missing layer. That argument is settled. What moved since the 2026-11-05 brief is the party doing the asking. Authorization stopped being a thing enterprises choose to govern and became a thing they must demonstrate to someone holding a deadline.
Two signals set the clock. SC Media lays out the December 2 EU AI Act milestone: a complete AI inventory covering system, provider, owner, data classification, and risk class. That is the non-human identity register described in regulatory language. Thirty days out. On the US side, Forbes counts the global rule inventory tripling from roughly 250 instruments in March to 619 in September, with more than 200 from US states. The author's framing is the part that matters: some of these require demonstrating specific system behaviors, not maintaining a policy binder.
Then the gap. The CUBE survey finds 59% of compliance teams say AI is embedded in core processes, 10% allow autonomous action, and only 34% are confident they could demonstrate AI governance to a regulator today. Run those numbers against the December deadline and the shape is clear. The deployment line has crossed the demonstrability line, and the demonstrability line is now law.
The 2026-W43 index scores Governance and Ethics at 79, the highest organizational dimension. Read it alongside the CUBE number and the 79 looks like policy investment, not operational confidence. Writing the policy was the easy half. Proving an agent did what the policy said, tied to a specific identity and action, is the audit surface almost nobody has built.
This is where today's tooling signals earn their place. Protegrity shipped a single policy engine governing human and non-human identities as peers, which is the enforcement layer the inventory requirement presupposes. Microsoft Research's PDD Kit moves the audit object from code to the specification behind it. Both point at the same operational demand: regulators want to inspect intent and action, and most governance frameworks have no register for either.
So the move for a principal this morning is narrow and dated. Pull your AI inventory and test it against the EU AI Act's five fields. If you cannot name the owner and risk class of every agent in production, you have a December problem, and it is a budget conversation before it is a compliance one.
Watch item: the first state enforcement action against an enterprise AI deployment. That is the trigger that moves board attention from policy to budget.
Index Reference · Applied AI Index 2026-W43, the latest published issue (Sunday, October 25, 2026)
WatchPost-midterm AI regulation: TechTarget reports analysts expect continued state-level AI regulation, with requirements varying by jurisdiction across disclosure, auditing, and oversight. Enterprises are advised to treat AI governance as a standing operational function. The Forbes signal in today's brief quantifies the trajectory: 619 instruments globally as of September. Watch for the first state enforcement action against an enterprise AI deployment as the trigger that moves board-level attention from policy to budget.