Jensen Huang’s AI capex question, applied to the work you finish.
OMB told agencies to add new LLM contract requirements to existing contracts before they extend them. The vendor's documents are in. The decision is not.
Joint Commission's new certification for responsible AI asks which deployment decisions needed governance approval and how those decisions were tracked. A monthly agenda slot is not an answer.
Europe's draft GMP annex on AI puts every change to a model under change control, and it asks for a full justification whenever someone decides not to retest. That decision needs an owner who is available.
FinCEN and the federal banking agencies said there is no requirement to document a decision not to file a SAR, and that a short statement will usually do. Someone still has to own the call.
Colorado's governance rule for insurer data and models now covers auto and health insurers, and it asks for delegated decision-making authority. A quarterly agenda slot is not delegation.
The Department of War's AI strategy says a denied data request must be justified within seven days. A request that just sits is not one of the options.
The D.C. Court of Appeals struck a brief that cited four cases that do not exist. Verification needs a named owner who can sign off before the deadline.
OMB's AI acquisition memo asks agencies to check new AI features before they go live. That check needs a named owner and a deadline.
Section 1557 gives covered hospitals an ongoing duty to identify and mitigate discrimination risk in clinical decision support tools. Finding the tool is the easy part.
Revised Fed, OCC and FDIC guidance recognizes that a model may need to go into use before validation is finished, with limits. Someone has to own that call.
FDA's draft guidance on AI in drug development turns model credibility into a decision someone has to own.
The NAIC model bulletin asks insurers to govern AI systems that support regulated decisions. Governance that waits is still a wait.
ABA Formal Opinion 512 put generative AI under the rules lawyers already follow. The slow part is deciding who answers.
CMMC Phase 1 makes a current status in SPRS a condition of award. The last step is a person.
OSFI's Guideline E-23 ties approval authority to model risk. Use it to stop sending every change to the top.
Microsoft's CEO says companies pay for intelligence once in money and again in know-how. Here is how regulated firms keep the second payment in-house.
Enterprise AI
+2
US bank regulators set one lifecycle for third-party risk. Where a tool runs and where the data goes shape how long that lifecycle takes.
Third-Party Risk
A 90-day test for turning disconnected experiments into measurable operating results.
+ useful autonomy, mandatory approval, and a hard stop
How to build a decision library that teaches new hires without turning yesterday's judgment into today's rule.
Every team had the latest plan, but her evening disappeared into discovering that they did not mean the same one.
Faster answers are useful only when a team knows which decision the answers are meant to support.
+ governance, safe failure, human authority
The CEO question is shifting from “Which model?” to “Who can release the decision?”
