Risk Belongs in the Workflow

OpenAI says it did not disband its preparedness team. Fine. Take the denial seriously.

The more useful question is not what happened to one team inside OpenAI. It is where risk work lives once AI leaves the lab and starts touching sales notes, analyst models, support replies, hiring screens, customer data, and the half-documented processes people actually use to get through the week.

If risk lives in a committee, it loses to the workflow.

That is not because committees are useless. Standards matter. Tool rules matter. Data rules matter. But work does not wait for a monthly review. Someone adds an AI step because the queue is backed up. Someone pastes source material because the deadline is real. Someone trusts an output because it looks clean enough and nobody built a better checkpoint.

The control has to sit where the handoff happens.

Pick one workflow where AI is already being used. Map the path. Where does information enter? Where does the model touch it? Who reviews the output? What breaks if it is wrong? Who owns the fix?

If nobody can answer that without pointing to a PDF, you do not have AI governance.

You have paperwork.