The Independence Problem
Model risk governance depends on independent challenge. Independent challenge depends on people who hold views formed outside the model. That is harder to guarantee when the risk team has spent years doing its thinking with AI assistance.
The independence does not disappear on paper. It disappears in practice. A reviewer who has internalised the model's framing, absorbed its assumptions, and lost the habit of reasoning from first principles will still sign the attestation. The oversight is real in form and hollow in substance.
2008 Did Not Happen Because Risk Teams Were Negligent
The risk functions that missed the build-up to the financial crisis were, in many cases, technically compliant. They had oversight structures, challenge processes, and governance frameworks in place. What they lacked was genuine understanding of the models those frameworks were supposed to govern.
AI adoption in risk functions does not solve that problem. It expands the surface area of it. The models are more capable, the outputs are more persuasive, and the gap between formal compliance and real understanding is wider than before.
What Cognitive Sovereignty Means for Risk Leadership
Steve works with risk leaders on the cognitive dimension of model risk governance. That means being precise about what genuine independent challenge actually requires from the people doing it, and honest about whether current practices build that capacity or quietly erode it.
The goal is not to use fewer AI tools. It is to build risk cultures where AI tools are used without accumulating the kind of dependency that makes oversight nominal. That distinction matters for regulators, for boards, and for the next time the models are wrong about something important.
Read the first chapter free
Steve's book, Cognitive Sovereignty, covers this in full. The first chapter takes about 20 minutes to read and is free.
Work with Steve
Steve speaks and consults with organizations working through exactly these challenges. See the Work with Me page for details.